Post on 08-May-2018
transcript
NOTICES
Product Information
Lecture Manual for: ACE Analyst
Software Release: 16.0.T PL0
Technical Support
Telephone: 240.497.1200
Facsimile: 240.497.1064
E-mail: support@opnet.com
Web: http://www.opnet.com
OPNET Technologies, Inc. (“OPNET”) is the sole owner of all proprietary rights in this Lab Manual (the “Lab Manual”) and the OPNET IT Guru software (the “OPNET Software”).
Terms and Conditions of Use
Access to and use of the Training Manual and/or the OPNET Software is restricted to authorized users who are employees or direct contractors of a Licensee who has accepted the terms and conditions of OPNET’s Software Master License Agreement. As a condition of having access to or use of the Lab Manual and/or the OPNET Software all users agree to be subject the terms of OPNET’s Software Master License Agreement. Users shall not copy or reproduce, in whole or in part, or permit access to or disclose the Lab Manual or the OPNET Software to any other person without the prior written consent of OPNET.
Liability Statement
The terms and conditions governing the licensing of OPNET Software consist of those set forth in the written contracts and license agreements between OPNET and its customers. No representations in this Lab Manual or statements of capability, suitability for use, accuracy or performance, shall be a warranty by OPNET unless contained in a written agreement signed by OPNET.
OPNET DISCLAIMS ALL EXPRESS OR IMPLIED WARRANTIES AS TO THE LAB MANUAL AND THE OPNET SOFTWARE INCLUDING IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
Destination Control Statement
This Lab Manual and the OPNET Software were manufactured in the United States by OPNET. The initial export of this Lab Manual and the OPNET Software, and any subsequent relocation to another country shall comply with United States law relating to the export of technical data, equipment, software and know-how.
Restricted Rights Legend
USE BY AGENCIES OF THE UNITED STATES GOVERNMENT: For purposes of use of this Lab Manual and OPNET Software by agencies of the United States government (or those authorized to use of this Lab Manual and OPNET Software on behalf of an agency), the OPNET Software is “Commercial Computer Software” within the meaning of DFARs 227.7202 and 252.227-7014(a)(1) and FARs 52.227-19. Use, reproduction and disclosure is subject to DFAR and FAR provisions relating to Commercial Computer Software and the License Agreement between OPNET and the agencies using OPNET Software. OPNET software is provided with a RESTRICTED RIGHTS NOTICE for government use.
Trademarks
OPNET, OPNET Technologies, Inc., OPNET Modeler, IT Guru, OPNET Development Kit, NetBiz and IT Guru are trademarks of OPNET Technologies, Inc. All other trademarks used in the Lab Manual are the property of their respective owners and used herein for identification purposes only.
CONFIDENTIAL INFORMATION DO NOT DISCLOSE, FORWARD, DISTRIBUTE, SHARE, OR MAKE COPIES OF THIS DOCUMENT IN WHOLE OR IN PART. This document contains confidential information and may contain information that is proprietary, privileged, and/or exempt from disclosure under applicable law. This document is intended for the exclusive use of the person to whom it is disclosed. If you are an unauthorized person, you are hereby notified that any viewing, copying, disclosure or distribution of this information may be subject to legal action. All unauthorized persons must immediately destroy the original documentation without making any copies or further unauthorized disclosure.
NOTICES
Product Information
Lecture Manual for: ACE Analyst
Software Release: 16.0.T PL0
Technical Support
Telephone: 240.497.1200
Facsimile: 240.497.1064
E-mail: support@opnet.com
Web: http://www.opnet.com
OPNET Technologies, Inc. (“OPNET”) is the sole owner of all proprietary rights in this Lab Manual (the “Lab Manual”) and the OPNET IT Guru software (the “OPNET Software”).
Terms and Conditions of Use
Access to and use of the Training Manual and/or the OPNET Software is restricted to authorized users who are employees or direct contractors of a Licensee who has accepted the terms and conditions of OPNET’s Software Master License Agreement. As a condition of having access to or use of the Lab Manual and/or the OPNET Software all users agree to be subject the terms of OPNET’s Software Master License Agreement. Users shall not copy or reproduce, in whole or in part, or permit access to or disclose the Lab Manual or the OPNET Software to any other person without the prior written consent of OPNET.
Liability Statement
The terms and conditions governing the licensing of OPNET Software consist of those set forth in the written contracts and license agreements between OPNET and its customers. No representations in this Lab Manual or statements of capability, suitability for use, accuracy or performance, shall be a warranty by OPNET unless contained in a written agreement signed by OPNET.
OPNET DISCLAIMS ALL EXPRESS OR IMPLIED WARRANTIES AS TO THE LAB MANUAL AND THE OPNET SOFTWARE INCLUDING IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
Destination Control Statement
This Lab Manual and the OPNET Software were manufactured in the United States by OPNET. The initial export of this Lab Manual and the OPNET Software, and any subsequent relocation to another country shall comply with United States law relating to the export of technical data, equipment, software and know-how.
Restricted Rights Legend
USE BY AGENCIES OF THE UNITED STATES GOVERNMENT: For purposes of use of this Lab Manual and OPNET Software by agencies of the United States government (or those authorized to use of this Lab Manual and OPNET Software on behalf of an agency), the OPNET Software is “Commercial Computer Software” within the meaning of DFARs 227.7202 and 252.227-7014(a)(1) and FARs 52.227-19. Use, reproduction and disclosure is subject to DFAR and FAR provisions relating to Commercial Computer Software and the License Agreement between OPNET and the agencies using OPNET Software. OPNET software is provided with a RESTRICTED RIGHTS NOTICE for government use.
Trademarks
OPNET, OPNET Technologies, Inc., OPNET Modeler, IT Guru, OPNET Development Kit, NetBiz and IT Guru are trademarks of OPNET Technologies, Inc. All other trademarks used in the Lab Manual are the property of their respective owners and used herein for identification purposes only.
CONFIDENTIAL INFORMATION DO NOT DISCLOSE, FORWARD, DISTRIBUTE, SHARE, OR MAKE COPIES OF THIS DOCUMENT IN WHOLE OR IN PART. This document contains confidential information and may contain information that is proprietary, privileged, and/or exempt from disclosure under applicable law. This document is intended for the exclusive use of the person to whom it is disclosed. If you are an unauthorized person, you are hereby notified that any viewing, copying, disclosure or distribution of this information may be subject to legal action. All unauthorized persons must immediately destroy the original documentation without making any copies or further unauthorized disclosure.
2/10/2011
1
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Introduction to AppTransaction Xpert (ACE Analyst)
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Day 1
Corporate Overview• Founded in 1986• Publicly traded (NASDAQ: OPNT)• HQ in Bethesda, MD• Approximately 600 employees• Worldwide presence through direct offices and channel
About OPNET Technologies, Inc.®
partners• Global distribution agreement with Cisco
Best-in-Class Solutions and Services• Application Performance Management• Network Performance Management• Network R&D
Strong Financial Track Record• Long history of profitability
test
g y p y• Trailing 12-month revenue of over $100M• Approximately 25% of revenue re-invested in R&D
Broad Customer Base• Corporate Enterprises• Government Agencies/DoD• Service Providers• Network Equipment Manufacturers
3 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
2
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Customers (partial list)
Energy / UtilitiesGovernmentFinancial Services Insurance
Healthcare / Bio / Pharmaceutical Manufacturing Media / Communications Services
test
Retail / Consumer Products Transportation / Logistics
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.4 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
OPNET’s APM Xpert Suite
End-user experience and network monitoring and analytics
Deep application component monitoring and analytics
Deep transaction tracing, analysis, and prediction
Passive database performance monitoring
test
5 OPNET Confidential – Not for release to third parties
© 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Network and server discovery and monitoring
“What's in your Application?”
2/10/2011
3
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
OPNET is an IACET Authorized Provider
OPNET Technologies, Inc. has been approved as an Authorized Provider by the International Association for Continuing Education and Training (IACET), 1760 Old Meadow Road, Suite 500, McLean, VA 22102. In obtaining this approval, OPNET has demonstrated that it complies with the ANSI/IACET 1-2007 Standard which ishas demonstrated that it complies with the ANSI/IACET 1 2007 Standard which is internationally recognized as the Standard for training organizations.
As a result of their Authorized Provider status OPNET is authorized to offer IACET Continuing Education Units (CEUs) for its programs that qualify under the ANSI/IACET 1-2007 Standard.
test
6 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Continuing Education Units
What is IACET?
IACET is a Standards Developing Organization accredited by the American Standards Institute (ANSI) and is an authorizing body for education providers that meet the American National Standard for continuing education and training (ANSI/IACET 1-2007 Standard). The IACET standard is what learners seek for quality and is the core of thousands of educational programs worldwide. (see: www.IACET.org)
What is an Authorized Provider?
The IACET Authorized Provider (AP) status means that an organization adheres to IACET’s internationally recognized ANSI/IACET 1-2007 Standard for Continuing Education and Training. Only Authorized Providers may award IACET Continuing Education Units (CEUs).
Wh t i CEU?test
What is a CEU?
The Continuing Education Unit (CEU) was created by IACET as a measurement of continuing education. One (1) IACET CEU is equal to ten (10) contact hours of participation in an organized continuing education experience under responsible sponsorship, capable direction, and qualified instruction.
7 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
4
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Earning CEUs for this Course
Introduction to AppTransaction Xpert – This course is worth 1.3 CEUs
• To receive these CEUs, you must– Attend the entire course
– Fill out the feedback form
– Score 80% in a test at the end of class
• Your test is scored in the next 10 business days
• If successful, you will receive a certificate denoting your CEUs in the mail within 2 weeks
test
within 2 weeks
• Your completion of the course is stored in our database for later retrieval as needed
8 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Let’s get to know each other…
• Name ?
• What company you work for ?
• Which Department ?
• What experience do you have with OPNET products?
• Did you know we changed the names… !!??
test
9 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
5
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Course Structure
Day 1
Introduction to AppTransaction Xpert
Day 2
Advanced AppTransaction Xpert
•About OPNET
•Visualizations in AppTransaction Xpert
•Troubleshooting application performance
•Predicting application behavior before deployment
•Advanced Topics
•Understanding Delay Components
•Parallel Effects, User think Time
•Troubleshooting TCP applications
test
deployment
•Capture and import strategies
•WAN Acceleration
10 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Class Format
• In general, four phase instruction
– Short lecture sections
– Demonstrations
– Related hands-on labs
– Discussion of lab applications
• Class is interactive – don’t hesitate to ask questions
• Please fill out the feedback form online @ eval.opnet.com
test– Give us information on your current network and application concerns
– Provide suggestions on how we can improve our products
– Let us know how we can improve this training class
11 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
6
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Agenda Introduction
- Licensing Options
- Overview, Use Cases
- Basic Workflow
Using AppTransaction Xpert to visualize applications- AppTransaction Xpert Visualizations
- Lab 1: Visualizing Application
Using AppTransaction Xpert to diagnose problematic applications- AppDoctor
Reporting capabilities of AppTransaction Xpert
R ttest
- Reports- Lab 2: Application Performance Troubleshooting
Using AppTransaction Xpert to perform performance predictive studies - QuickPredict and QuickRecode
- Lab 3: Predictive Analysis
12 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Licensing Options
• Download and administer licenses using the License Manager
• Licensed on concurrent user basis
• Online license management g(http://www.opnet.com/support/license_management.html)
• Licensing options
– Standalone vs. Floating (license server)
– Loanable Licenses
test
13 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
7
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Overview of AppTransaction Xpert
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
AppTransaction Xpert Solution
• AppTransaction Xpert Plus
• AppTransaction Xpert Standard
• IT Guru with the following modules
– ACE
– ACE Decode Module
– Advanced ACE Capabilities
• Starting with version 16.0, the different offerings of AppTransaction Xpert (that is: AppTransaction Xpert Plus, AppTransaction Xpert Standard, and AppTransaction Xpert Module) have been combined into a single installer.
test
Important—Existing users are not losing any AppTransaction Xpert functionality
because of this change. Functionality is determined by licenses.
15 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
8
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
What is AppTransaction Xpert... ?• Application Characterization Environment
• Capture application traces
Vi li A li ti D i• Visualize Application Dynamics– Application layer analysis– Network layer analysis
• Diagnose performance problems– Minimize troubleshooting time
test• Validate solutions
– Predict application behavior under different network conditions– Evaluate performance impact of application design decisions– Validate infrastructure readiness for new deployments
• Generate reports16 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
When to use AppTransaction Xpert ??
(Use Cases)test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
9
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Use Case 1
Troubleshooting application performance –
• Users of a critical application are reporting very bad response times
– Network engineers are reporting no issues on the network
– Application engineers are reporting no issues with software
• Result: Finger pointing !!
Wasted time
test
18 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Use Case 2
Testing changes of the enterprise environment –
• Relocation of key processing data center from US to China
• Management needs to know the implications of move
– Impact on application performance and network utilization?
– Will additional resources be required?
• Result: Need to produce accurate performance projections and network l i i
testplanning requirements
19 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
10
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Use Case 3
Application pre-deployment analysis –
• Purchase of a new 3-party application system
• Management needs to know how the new system will perform
– Application response times
– Network utilization
• Result: Need to eliminate the need for trial and error deployment of
testapplication. Expensive and time consuming.
20 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Basic Workflow
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
11
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
What does AppTransaction Xpert tell you … ?
• “Why is my application slow?”
– User’s desktop running slow?
– Server delays?y
– Non-optimal protocol settings (TCP and others)?
– Heavy congestion over WAN?
– Too little bandwidth?
– Application too chatty?
• “Can this new application be deployed over our network?”
test– Expected transaction response times?
– Impact other critical applications?
– Bandwidth requirements?
22 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
How does AppTransaction Xpert tell you this...?
• Reports, graphs, and tables
• Visualization of application behaviorpp
• Diagnosis of bottlenecks and performance problems
• Predictive analysis of same traffic on different network
• Input to other OPNET products for further analysis
test– Application traffic analysis
– Capacity planning studies
23 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
12
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Inputs to AppTransaction Xpert
• Packet traces from the following sources– ACE Capture Agents
ACE Live– ACE Live– Cisco NAM Blade– WAN Accelerator– F5 BIG-IP– Sniffer– UNIX (no agent installed)– Tcpdump, windump ..etc
test• Basic characteristics of network (Recommended)
– Tier locations– Bandwidth between tiers– Latency between tiers
24 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Application Troubleshooting Workflow
Identify performance issues
Capture Packet Traces
Optional: Create IT Guru
Visualize, Diagnose, and Predict
Optional: Detailedtest
Optional: Create IT Guru network model
Generate Reports
Optional: Detailed Analysis
25 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
13
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
AppTransaction Xpert Visualizations
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
AppTransaction Xpert Visualizations
• Data Exchange Chart– Shows timeline of data sent
• Tier Pair CircleTier Pair Circle– Shows summary statistics
• Tree View– Shows Transaction Analysis
• Decodes– Shows Application layer headers
test• Graphs– View time-based graphs of application statistics
27 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
14
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Data Exchange Chart (DEC)
• Timeline of application messages and network packets between tiers
• Slope indicates time to cross network
• Color indicates payload size of message
test
28 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
DEC: Message Groups
• When messages and packet are too frequent to be displayed individually, a solid block is shown
• Block is a histogram showing the size of messages in groups
• Zoom in or use tooltip to see more detail
test
29 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
15
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
DEC: View Options•Application Chart Only
Shows every application message – several packets may comprise a message
• Network Chart Only
Shows every individual packet in the trace file, including TCP acknowledgements
• Application and Network Charts
Shows both charts in the same window
testApplication Messages
Network Packets
30 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
DEC: Show Dependencies
• Triangles appear at largestchunks of delay
Click “Show Dependencies”
• White = delay at host
• Orange = user think time
• Purple = network delay
test
LegendLegend
31 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
16
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
DEC: Application Characteristics
• Quiet Time
– User think time?
– Slow client computer?p
– Slow server?
– Not seeing the whole picture?
§Tip: See what happened before and after silence
test
f
Tip: Check Network Chart for more details
Large periods of silence
32 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
DEC: Application Characteristics (cont.)
• Chatty Application
– Lots of back-and-forth communication
(Application Turns)
Lots of small messages (orange and yellow)
test
Groups of messages close together
33 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
17
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Application Turn
• Counted on the Application Layer (Application Messages)
• Every turn has to cross the network latency which adds up to the total Latency Delay
2
Client
Networktest
1 3
Server
NetworkLatency
34 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
DEC: Application Characteristics
• One-way data transfers– Dark blue arrows
Most arrows in same direction– Most arrows in same direction
Tip: Check the Network Packet Chart for more details
test
Zoom to see the arrow heads
35 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
18
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
DEC: Application Characteristics (cont.)
• Lots of irrelevant traffic– Do not try diagnose
application untilapplication until you removed irrelevant data
test
Tip: Go to Tier Pair Circle and delete unrelated traffic
36 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Tier Pair Circle
• Summary of traffic between all tiers
– A high-level view of theg
application task
– Is there extraneous traffic?• Right-click to exclude
unwanted traffic
test
Tip: If you don’t see the protocol you expect, then right-click to re-recode traffic on the fly
37 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
19
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Tier Pair Circle: Functionality
• Choose statistics to view
• Right-click a tier or conversation to exclude /delete it or all others
• Show only traffic containing a specific protocol
• Find Top Statistics to sort test
pconversations by relevance
38 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Tier Pair Circle: Statistics
• Pull down to select statistics to view:
test
39 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
20
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Tree View
• Tree pane: One-line summary of tier pairs, connections and frames
• Table pane: Detail statistics by tier pairs, connections and frames
• Timeline pane: Overview of messages over the time of the taskp g
test
Tree Pane Table Pane Timeline Pane
40 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Tree View: “View by:” Options
• Tier Pairs – Network Packets
• Tier Pairs – Application Transactions
• Application Transactions
Group by:
test
41 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
21
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Tree View: Sorting
• Click a column header to sort the table by descending values.
• Click the table header again to sort by ascending values.
test
42 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Tree View : Customize the Table Pane
• Choose View > Choose Table Columns…
• Right-click on the column header and click “Choose TreeView Columns…”
• Select the network statistics to appear in the table
test
43 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
22
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Decode Viewer
ETH IP TCP HTTP ETH
• What are decodes?
– Analysis or description of the bytes at each layer of a packet
test• What can you learn from decodes?Detailed protocol information
Information about data being transferred
Ex: TCP gives Sequence number and Acknowledgement number
44 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Embedded Decode Viewer
• Show / Hide the embedded decode viewer by clicking and
• Select traffic in the top pane and see corresponding decodes in the bottom pane
• Right-click on selected
testframes to export to spreadsheet
• Click the plus sign to expand layers of decodes
45 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
23
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Tree View to DEC
• Display Context-Based Traffic in Data Exchange Chart —
– Double-click on traffic for a tier pair in the Timeline pane to display the corresponding context- based traffic in the Data Exchange Chart.
test
46 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Graphs
• Time varying statistics
– Advanced diagnosis of network and TCP problems
test
47 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
24
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Graphs (cont.)
• Select graphs to display
• Options:
– Specify bucket width for precision of graph
– Embed graphs in DEC for easy comparison against messages in DEC
test– Overlay similar graphs or overlay all
• Select “Show”
48 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Graphs: Bucket Width
• What is a bucket width?
– Statistics can be averaged over time rather than shown as individual data points
– A bucket is the duration of data you want to average, like 100 ms or 3 seconds
• Same trace, same statistics, different bucket widths:
test
d e e t buc et w dt s:
– 100 ms
– 3000 ms (3 seconds)
49 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
25
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Graphs: Embedded Statistics
• Time-aligned statistics with the Data Exchange Chart
test
50 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
LAB 1
LAB 1
Visualizing
Application
test
51 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
26
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Lab 1: Visualizing Application
• Overview
– The ACE trace file represents a three-tier Oracle application. You will p ppexplore some of the different visualizations in AppTransaction Xpert that you can use to troubleshoot and analyze the application.
• Starts on page 1 in your lab manual
test
52 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Lab 1: Conclusion
AppTransaction Xpert provides many different views of a transactionto understand the interaction of the application on the network
• Data Exchange ChartData Exchange Chart– Shows timeline of data sent
• Tier Pair Circle– Shows who talks to whom and summary statistics
• Tree View– View the network packets in order, with a decode summary– Transaction analysis
• See the timeline of transactions• Decodes
test
Decodes– Drill down into TCP, IP, or even application layer headers to understand
application behavior• Graphs
– View time-based graphs of throughput, congestion, and other statistics
53 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
27
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
AppDoctor
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
What is AppDoctor … ?
• Invoke AppDoctor– AppDoctor (menu)– AppDoctor (icon)
• AppDoctor Tabs
– Summary of Delays
– Executive Summary
– Diagnosis
– Statistics
– Protocol-specific
test
55 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
28
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
AppDoctor: Summary of Delays
• Pie chart that breaks down the total response time into processing, network, and
ll l ff tparallel effects
• Most useful view in AppTransaction Xpert when troubleshooting applications
• Quickly points you to what is
testthe problem, or at least what is definitely not the problem
• Must be used on a single application task
56 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Summary of Delays: Processing Effects
• Tier Processing delay
– Time spent processing on each tier
test
57 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
29
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Summary of Delays: Network Effects
• Latency– Delay caused by the physical
distance between tiers– Common problem in chatty p y
applications• Bandwidth
– Delay caused by transmission (i.e. putting the bits onto the wire)
• Protocol – Could be non-optimal TCP
settings – Check Diagnosis tab for
test
Check Diagnosis tab for bottlenecks
– Check Statistics tab for values• Congestion
– Delay caused by other traffic in the network
58 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Summary of Delays: Parallel Effects• Delay incurred when multiple types of delay (such as tier processing,
bandwidth, and latency) occur simultaneously
test
59 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
30
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
AppDoctor – After Network Definition (16.0 )
test
BEFORE AFTER
60 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Summary of Delays: Drill-Downs
• Double-click any section of the pie chart for extra information
• Drill-downs give more information that can help diagnose problems
• Ex: Bandwidth drill-down shows throughput graphs and link bandwidthg p g p
test
61 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
31
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
AppDoctor: Executive Summary
• Gives overview information for application task– Total transaction time– Total application data transferred
Si l t t f d l– Single greatest cause of delay– Tiers in application– Bandwidth and latency at each tier
test
62 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
AppDoctor: Diagnosis• Displays common
application, network, and server problems
• Flags bottlenecks and• Flags bottlenecks and potential bottlenecks
• Various characteristics of the ACE file are compared with thresholds
• Click on any cell
test
Click on any cell (especially on bottlenecks) for extra information
63 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
32
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
AppDoctor: Statistics
• Tabular view of key application characteristics
• “Effect of” statistics quantify• Effect of statistics quantify the components of delay
• Other statistics include– Application message statistics
– Network packet statistics
– TCP statistics
testTip: Use it with Diagnosis tab to quantify bottlenecks
Tip: You can export to spreadsheet
64 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
AppDoctor: Protocol-Specific Tabs
• Protocol-specific information, statistics, and diagnoses
• Appear automatically when certain protocols are detected in an ACE file:
– HTTP
– Citrix
– CORBA
– SOAP (Web Services)test
SOAP (Web Services)
65 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
33
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Right-Click Operations in AppTransaction Xpert
• Graph Statistics such as throughput or congestion
• Exclude or permanently delete packets and messages
• Include packets and messagesp g
• Rename a tier or lookup a tier name with DNS
• Re-code traffic
• Zoom in and out in the DEC
• Show protocol decodes for selected traffic
• Export decodes for selected frames in the decode viewer
test
66 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Take-Away Points
AppTransaction Xpert provides an analysis of the main application bottlenecks that is easy to use and to understand. The diagnosis is displayed and broken down into components of delay. The different components are:
• Processing Effects
• Network Effects– Latency– Bandwidth– Protocol (with recommended options)
C titest
– Congestion
• Other Effects– Parallel Effects: Combination of any processing effects and network
effects that occur at the same time
67 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
34
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Reports
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Reporting Motivation
• Quick and easy to use
• Support your findings
• Recording and dissemination of all analysis results to all interested stake holders
• Documentation of application behavior for future reference
test
69 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
35
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Report Selection - Type and Format
• Very simple and quick – select report type for output (from pull down menu, or icon button)
test
70 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Report Options
• Select report options and generate report– Report sections, title (statistics)– Report directory (output), Report Server (optional)
test
71 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
36
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
ACE – MS Word Report
test
72 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
ACE – Web Report URL
Home Dir(all reports)
test
73 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
37
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
ACE – Multiple Transaction Report
Trans 1
Trans 2
test
74 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
ACE – Difference Report
test
75 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
38
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
ACE – Spreadsheet Export Reports
Comparison Report
Tree View Statistics
AppDoctor Statistics
test
76 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
• Report template — Defines the previously-defined templates used
to generate the report. You can modify and customize report templates.
The standard templates:
ACE – Report Template
– ApplicationDeveloperReport.ace.rt.txt
– ManagementReport.ace.rt.txt
– NetworkEngineerReport.ace.rt.txt
– concise_report.ace.rt.txt
– detailed_report_ace.rt.txt
– standard_report_ace.rt.txt
• Reports can be cataloged and managed by the Report Servertest
• Reports can be cataloged and managed by the Report Server– Provides security, allows access by only registered users
– Provides automatic ageing of reports
77 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
39
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
LAB 2
LAB 2
Application
Performance
Troubleshootingtest
78 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Lab 2: Application Performance Troubleshooting
• Overview– Response time problems for remote users of new web application
– Remote site connected via a 256-Kbps Frame Relay circuit
– Need to determine cause of the delay (network or web server)
– Traces have been captured at both server and client
• Objectives– Visualize application in ACE
– Identify the cause of the application delay
– Use Transaction Analysis to identify problem with application
test• Starts on page 9 in your lab manual
79 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
40
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Lab 2: Conclusion
• Conclusions
– The web page takes approximately 24 seconds to download
– Bandwidth and congestion are the biggest sources of delay
– Server delay is not a problem
– Majority of traffic flow is from the server to the client
– Web page consumes almost all of the 256-Kbps circuit bandwidth
– Largest transaction is a download of a 352 KB graphic
test– This single graphic takes 18 seconds to download
80 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Performance Predictive Studies
The use cases for predictive studies:
• Predict application behavior under different network conditionspp– Problematic applications
• Evaluate performance impact of application design decisions – Modified applications– Problematic application
• Validate infrastructure readiness for new deploymentstest
Validate infrastructure readiness for new deployments
81 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
41
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Prediction in AppTransaction Xpert
• What can predictive analysis tell you?
– Response time of application under different circumstances
– Will adding bandwidth help?g p
– Is this application sensitive to latency?
– Can we roll this out to remote users?
• What sort of studies will it help with?
– Troubleshooting performance problems
– Testing a new application
test– Deploying a current application to new users
– Server relocation studies
82 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Application Deployment Workflow
Identify performance issues
Capture and import into ACE
Create report and prediction scenario templates
Use QuickPredict to predict performance
O ti l U M lti Utest
Optional: Use Multi-User QuickPredict for basic capacity
planning
Optional: Use IT Guru for network and server capacity
planning83 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
42
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Application Deployment Workflow
• Optional: Use simulation for further studies
– Discrete Event Simulation to predict response times or perform server capacity planning
– Flow Analysis for network capacity planning
– Simulation covered later in course
test
84 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
QuickPredict
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
43
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
QuickPredict
• Simplified simulation engine
• Helps study effects of the network on the application
• Questions QuickPredict can answerQ Q
– How much bandwidth would it take to get a 4 second response time?
– Would increasing the TCP Window improve response time?
– What happens to response time if latency increases?
– How sensitive is this application to packet loss?
– What would the response time be if the link was 65% utilized by other traffic?
h if h d 10% A A l i i k !!test
– What if I had a 10% WAN Acceleration in my Network – NEW !!
86 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
QuickPredict: Outputs• Two different output styles
– QuickPredict Bar Charts shows components of delay for different scenarios
– QuickPredict plots response time over range of network conditions
test
87 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
44
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
QuickPredict Bar Charts: Scenarios
• Click to add multiple scenarios
• Right-click to remove scenarios
test
88 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
QuickPredict Bar Charts: Options
• Put parameters on slider
– Drag slider back and forth to dynamically update bar chart
• Rename scenarios
• Symmetric Bandwidth
– Set asymmetric values for bandwidth
• Label delay categories
• Load and save parameter settings using templates
test
using templates
89 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
45
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
QuickPredict Bar Charts: Options – NEW
Introduce WAN Acceleration• Byte Reduction • Turn Reduction
C i• Compression • Caching
test
90 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
QuickPredict Bar Charts: Tabular Results
• View detailed results for each scenario in the Tabular Results tab
test
91 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
46
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
QuickPredict: Response Time Graphs
• Shows sensitivity of task response time to network characteristics:
– Bandwidth
– Latencyy
– Packet Loss
– Link Utilization
– TCP Window Size
– WAN Acceleration
test
92 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
QuickPredict: Advanced View
• Choose network path to modify
• Choose an X-axis parameter and the range of values over which to plot response times
• Select values for each of the other parameters
• Click Update Graph
• Overlay multiple curves on same graph
• Load and save templates
test
93 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
47
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
QuickPredict: Table View
• Modify parameters on all network paths in the task
–Red text indicates X-axis parameter–Use Set X-axis to specify X-axis parameterp y p–Load and save templates
test
94 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
QuickRecode
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
48
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
QuickRecode
• Changes high level application characteristics– Shows sensitivity of task response time to application characteristics:
•Application turns•Total application bytes sent•Total processing time in tiers
– Use with QuickPredict to see how response times are affected
test
96 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
QuickRecode: Use Cases
• Use Cases– Performance benefits– Changing database access
to reduce chattiness– Demonstrate effect of faster
app serverHint: Use QuickRecode to
prove effect of optimizing application
test
97 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
49
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
LAB 3
LAB 3
Predictive
Analysis
test
98 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Lab 3: Predictive Analysis
• Overview– Deploying a new application to several remote locations– Application has client querying an Oracle database– Application has been deployed to pilot site (in same city as database) and
poor performance is already apparent– Goal is to determine if the application deployment will be successful– If not, what can be done to improve performance?
• Objectives– Perform a deployment study using QuickPredict Bar Charts– Use AppTransaction Xpert to drill down into the app and see what’s going
teston
– Use QuickRecode to test a possible fix
• Starts on page 20 in your lab manual
99 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
50
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Lab 3: Conclusion
• Always try to determine “goal” or “objective” of any problematic application– Is there an SLA expectation– What is the user expectations– What are other users, departments, experiencing
• Additional useful troubleshooting information to determine– Is the problem reproducible or is it transient– Are all transactions of the application behaving the same
test
100 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Agenda Capture Methodologies
Capture Data Sources
- New Capture Manager Options
Using AppTransaction Xpert to perform packet capturesg pp p p p p- Capture Types
- Lab 4: Capture Process
Using AppTransaction Xpert to filter packet captures
- Capture Filters
Using AppTransaction Xpert to import packet captures- Import Process
L b 5 Id tif i P f P bl i U filt d Ttest
- Lab 5: Identifying Performance Problems in Unfiltered Traces
101 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
51
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Capture Methodologies
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Basic Capture Methodology
• Define and understand the problem at hand– What is slow?– Is it the login or another specific task?– How often does it occur?
• What is a task?– Isolate an “atomic” Transaction– Loading single web page– From when user clicks “Submit” to screen refresh
• Plan your capture– Simultaneous or single-sided– Which tiers are involved in the application?
test• Aim for as “clean” a packet trace as possible– Keep measurement as simple as possible– Filter out unrelated traffic– Capture one occurrence of one task at a time– Perform capture numerous times to help eliminate anomalies
103 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
52
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Atomic Transaction Example (Task)
• Web Application of a Bank– Load Login Page
• Open a Web Browser• Type URL of the Bank yp• “Enter”• Login Page Loads
– Login• Type User credentials • “Login”• Home Page shows up
– Check Savings Account• Mouse/cursor around• Click on “Savings Account”
Red arrow = Start Capture
Blue arrow = Stop Capture
test
Click on Savings Account• Savings Accounts Page Loads
– Logout• Call up Logout screen• “Submit”• Logged out page
104 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Capture Data Sources
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
53
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Capture Data Sources
• Packet traces are the primary data inputs
• AppTransaction Xpert supports a wide variety of packet sourcespp p pp y p
– ACE Capture Agents
– ACE Live
– ACE Enterprise Management Server
– Agentless Capture Options – NEW!
• F5 Load Balancers
• UNIX, Linux platforms
test• WAN Accelerators
• Cisco NAM
– Wireshark and other Winpcap and Libpcap sources
• TCPDump, Snoop
106 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Agentless Capture Options – NEW
• No software installation required
• Use the AppTransaction Xpert Capture Manger to control captures remotely via SSH
• Supported on platforms that support TCPDump– TCPDump file remains on machine until capture is completed, then deleted after
retrieval
• Included in the AppTransaction Xpert Plus license
test
107 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
54
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Agentless Capture Options – NEW (cont.)
• F5 Load Balancers– Excellent way to follow a transaction to the proper server– Perform successful trace file merging across firewalls
Rule out the device as a source of delay– Rule out the device as a source of delay
• UNIX-Based Operating Systems– Supported Platforms
• Linux• Solaris SPARC• Solaris Intel/AMD• AIX
test• FreeBSD• HP-UX Itanium• Macintosh PowerPC• Macintosh Intel• True64
108 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Agentless Capture Options – NEW (cont.)
• WAN Acceleration Devices
– Support for Riverbed and Cisco
– Restore visibility of the WAN
– More information can be found in session 1471 - Monitoring and Troubleshooting Applications in a WAN Accelerated Environment with ACE Live™ and AppTransaction Xpert
test
109 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
55
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Capture Data Sources – AppResponse Xpert (ACE Live)
• New seamless 1-click integration– Let ACE Live do the filtering for you
• Easily pull traces from alert snapshots f t l li ti iof actual application issues
test
110 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Capture Data Sources – AppForensics Xpert (ACE EMS)• ACE Enterprise Management Server
– Provides web-based extension to Continuous Capture functionality– Supports both End-User Driven Capture workflows, as well as Server-Driven
Capture WorkflowsCapture Workflows
test
– End Users report application slowness• Trouble ticket is created with trace files of the incident attached• Troubleshooters claim problem reports and troubleshoot on their schedule
111 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
56
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
AppForensics Xpert – Web Capture Retrieval
test
112 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Capture Data Sources – AppForensics Xpert
• Server-Driven Capture Workflow– Panorama provides forensic actions based on threshold violations within the
application servers
– The trace file forensic action automatically generates a trace that is accessedThe trace file forensic action automatically generates a trace that is accessed through AppForensics Xpert
– For more detail, attend session 1482 – Deploying an Enterprise-Wide Packet Capture Architecture with ACE EMS and ACE Live
test
113 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
57
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Capture Data Sources – ACE Agents
• Benefits of using ACE Agents– Unlicensed– Support for major platforms (Windows, HP-UX, AIX, Solaris, Linux, Mac OS X,
FreeBSD)FreeBSD)– Lightweight, Secure, and Stable– Silent Install capability– Server Statistics collection (Windows)– Enhanced Filtering (NEW) – On-Demand and Continuous Capture modes
test
114 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Additional Data Sources
• AppTransaction Xpert can import a variety of additional data– Identify these sources during capture to ensure collection
• From OPNET products– Windows ACE Capture Agents can collect Lightweight Server Statistics
– OPNET Panorama
• Server statistics and performance metrics
• Java transaction tracing
• Third-party sources– HP GlancePlus
– Perfmon
test– Custom data in commasaved (.csv) format
115 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
58
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Capture Lightweight Server Statistics
• Capture Windows performance data – ACE Capture Agent generates two files
– Application traffic (*.appcapture) file
– Windows performance data (*.appcapture.da) file
Display internal server metrics • Percent Processor Time•File Read Bytes Per Second•System Calls Per Second•Available MB
test
Available MB•File Write Bytes Per Second
116 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
AppInternals Xpert (Panorama) Server Statistics
• Capture Server Statistics from AppInternals Xpert – Use with on-demand captures
– Thousands of available metrics
• For more information:– Session 1461- OPNET Panorama® Essentials: Multi-Tier Application Monitoring,
Correlation, and Troubleshooting
test
117 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
59
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
AppInternals Xpert Transaction Trace
• AppInternals Xpert can create a trace of all activity through the Java Virtual Machine (JVM) or .NET CLR– Provides a call-by-call sequencing (or call stack) of all activity within your
li iapplication
• This transaction trace can be imported into AppTransaction Xpert to explain what is occurring on a given system during long processing delays
test
118 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Capture Manager
• Capture Manager control parameters
test
119 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
60
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Capture Agent Settings
test
120 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Capture Types
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
61
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
On-Demand Capture
•Methodology–User knows exactly when a problem is going to occur and
what it takes to reproduce itwhat it takes to reproduce it
–Deploy and configure ACE Capture Agents
–Start the capture
–User re-creates the problem
–Stop the capture
–Import capture for analysistest
Import capture for analysis
122 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
On-Demand Capture (cont.)
test
123 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
62
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
On Demand Capture Process: Adding an Agent
test
124 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
On Demand Capture Process: Specify Agent Settings
test
125 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
63
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Capture Process: Start Capture
test
126 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Capture Process: Finish Capture
test
127 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
64
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Capture Process: Open in Transaction Analyzer
test
128 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Scheduled Capture
•Useful for reproducible problems during off hours
•Methodology–User knows when a problem will occur
• For example, after-hours batch process or database backup
–Schedule a capture• Choose one-time or repeating capture
• Set the schedule for a capture
• Specify schedule for download
I t t f l itest
–Import trace for analysis
129 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
65
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Scheduled Capture (cont.)
test
130 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Scheduled Capture (cont.)
Scheduled capture can create one-time, daily, weekly, or monthly capture schedules
test
Scheduled capture utility can optionally create a one-time, daily, weekly, or monthly schedule to download trace files
131 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
66
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Continuous Capture
•Useful for troubleshooting intermittent problems
•Methodology–User experiences intermittent problem
–Start continuous capture
–User calls help desk• Problem occurred again around 9:30
–Preview capture
Download trace of problematic time spantest
–Download trace of problematic time span
–Import trace for analysis
•Agent details–Same capture agents as on-demand capture
132 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Continuous Capture (cont.)
test
133 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
67
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Continuous Capture: Preview/Download
test
134 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
LAB 4
LAB 4
Capture
Process
test
135 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
68
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Lab 4: Capture Process
• Overview
– Configure a capture agent
– Perform an on-demand capturep
– Open in Transaction Analyzer
• Skills covered
– Learn how to configure a capture agent
– Understand how to use agent filters
– Capturing from the network
test• Starts on page 32 in your lab manual
136 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Lab 4: Conclusion
• Configured capture agent for capture– Located and specified host IP Address
– Specified capture mode
– Set up a filter
• Performed a capture– Started the agent
– Performed a transaction
– Stopped the agent
– Brought the capture file into Transaction Analyzer
test
g p y
137 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
69
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Capture Filters
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Filtering Unrelated Traffic
• You want to capture the transaction of interest– AppTransaction Xpert lets you filter out irrelevant traffic
• Options to filter during capture, during the import or after the import– The earlier the better
• Why filter during capture?– Only capture the specific transaction– Smaller trace file to process => easier/faster import
• Why not filter during capture?– Sometimes you want to see the whole picture– If you’re not 100% sure about application path
test• Recommended methodology– Capture without filters for short duration– Examine trace– Determine best filter to use– Capture subsequent tasks with filter
139 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
70
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Filter: During Capture
Users can filter out or only allow traffic based on a specific protocol
Users can filter specific hosts and/or ports for incoming or outgoing traffic
New Filtering Options intest
Users can elect to exclude or allow broadcast and multicast traffic
New Filtering Options in Capture Agent 3.9 Use wildcards and slash notation Use DNS names Port ranges Broadcast and multicast can be
excluded independently140 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Filter: During Capture (cont.)
• Preview in Continuous Capture– Select the IP addresses, conversations and time ranges of interest
• Great for troubleshooting intermittent issues
test
141 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
71
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Filter: During Import
•What is Trace Explorer?– Graphical protocol analyzer
– Used to help filter unrelated traffic pduring import process
•Why would you use it?– Graphically filter a capture file
– Allows for fast read of large trace files
– You want to know what you’re looking at
test
looking at
142 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Filter: After Import
• Once capture file has been imported, users can still filter irrelevant traffic
– Right-click on any tier pair conversation, application message,
li i i kapplication transaction or network packet
• Option to Exclude or Permanently Delete
– Filter Wizard• Remove unidirectional, non-IP, UDP,
broadcast traffic
– Capture File Filter
test• Use to remove traffic by host, port,
protocol
143 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
72
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Import Process
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Import Overview
• Now that the data has been captured, it must be imported into AppTransaction Xpert
• Review what has been captured– Do you have a single capture file?
– Do you have more than one capture file?
– Are you capturing WAN accelerated traffic?
– Is the capture file filtered?
• If not, use Trace Explorer
test• Greatly simplified for 16.0 !!
145 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
73
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Import Changes in 16.0
• Import process has been greatly simplified– 15.0 required answering a series of questions about tier locations, bandwidths, and
latencies
– For 16.0, we focused on asking as few questions as possible• Filter, view statistics or diagnoses, or explore your transaction immediately
• Locations, bandwidths, and latencies can still be specified to break down Network Transfer effects into Bandwidth, Protocol, and Congestion
• Most importantly, you can change these values whenever you need to
• Import is now broken down into use cases– Single packet trace
– Merge several traces
test
Merge several traces
– Merge traces from a WAN-accelerated environment
– Batch import
– Filter into Trace Explorer
146 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Import a Single Capture File
test
147 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
74
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Merging Capture Files
• What is a merge?– Two or more traces of an application transaction can be combined
– Creates unified view of application
– Accurately determines the send and receive times of each packet
• Benefits and use cases– Allows analysis of multi-tier transactions where all traffic does not
go through one location
– Improves analysis in a production environment with congestion issues such as packet loss and retransmissions
test
148 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Merging Capture Files (cont.)
• All traces must be of the same transaction
• All must have been captured at the “same time”
• Easier to import if there is some common traffic between tracesp
test
149 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
75
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Merging Capture Files (cont.)
File name
test
Load previously saved import details (tier names, tier locations, bandwidth, latency,…)
150 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Merging Capture Files (cont.)
• Replace IP addresses with friendly names (can use DNS)
test
151 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
76
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
After Opening/Merging
• Now what?– Filter down to a single transaction or remove user think time
– Open AppDoctor
• Note that there are only Processing and Network effects identified
– Specify bandwidth and latency to break out individual network effects
test
152 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
AppDoctor
Bandwidth, protocol and congestion are all included in one category named “Network Transfer”
To break down the various t f “N t k
testcomponents of “Network Transfer”, click the “Refine Network Effects…” button
153 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
77
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Refine Network Effects
Specify the local and remote bandwidths here
test Tier locations and latencies are automatically detected, but can be changed if needed
154 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Specify Bandwidth and Latency
• Minimum bandwidth– Capacity of the slowest link
– Should be a known factor about your network
• Minimum one-way latency– PathProbe provides an estimate
– Half the value returned by ping
– Auto-calculated in a trace merge
– In many cases, AppTransaction Xpert can estimate latency of a single-sided
test
y , pp p y gtrace
155 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
78
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Specify Bandwidth and Latency (cont.)
• Quiz– Bandwidth of each segment is known
– Ping reports 46ms RTT
test• What would you specify for bandwidth and latency between the Client and the Server during import into AppTransaction Xpert ?
Bandwidth = 1544 Kbps Latency = 23ms
156 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
AppDoctor – After Network Definition
test
BEFORE AFTER
157 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
79
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
LAB 5
LAB 5
Identifying Performance Problems in Unfiltered Traces
testUnfiltered Traces
158 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Lab 5: Identifying Performance Problems in Unfiltered Traces
• Overview– Open a single side capture
– Filter traffic
– Refine network effects
– Analyze transaction
• Skills covered– Step through the import process in AppTransaction Xpert
– Understanding the importance of bandwidth settings
test
Understanding the importance of bandwidth settings
• Starts on page 42 in your lab manual
159 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
80
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Lab 5: Conclusion
• Imported a single-sided capture file– Imported capture file
– Filtered irrelevant traffic
– Estimated locations
– Entered local and remote bandwidth
– Estimated latency
test
160 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Introduction to AppTransaction Xpert
2test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Day 2
2/10/2011
81
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
What to Expect in Advanced AppTransaction Xpert Topics
• Mix of lecture and lab
• Advanced class– Address network concepts important to using AppTransaction Xpert
– Dive into “the science” behind ACE
– Highlight new ACE features
– Discuss TCP
• Please be interactivetest
Please be interactive– Participate
– Ask questions
162 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Prerequisites
• Prerequisites – Reviewed yesterdays material
– Reviewed session 1441
– Equivalent experience
• You should already understand the mechanics of ACE– Capturing a trace
– Importing traces
– Generating reports
test• You should already recognize “vanilla” problems– Chatty applications
– Insufficient bandwidth
– Slow servers
– …
163 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
82
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Agenda How AppTransaction Xpert Calculates and Predicts Components of Delay
- Lab 1: Calculating Network Delays
Understanding the different Components of Network Delay Understanding the different Components of Network Delay
- Congestion Delay
- Protocol Delay
- Lab 2: Understanding Components of Delay
Understanding Parallel Effects
- Parallel Effects
test
Parallel Effects
- User Think Time
- Lab 3: Understanding Parallel Effects
164 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
How AppTransaction Xpert calculates and predicts components of Delay
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
p y
2/10/2011
83
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Components of Application Delay
• The following are key components of delay in a transaction:– Processing
• Client
• Server(s)
– Network
• Protocol (TCP WIN, Nagle’s, etc.)
• Bandwidth
• Latency (baseline, dominated by network hardware and ultimately ruled by C)
• Congestion (including queuing)
• Packet loss
test– Application itself
• Think of the top two topics as “resources”. How does the app use them?– Efficient/inefficient code
– App turns/Chattiness
166 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Review of Network Delays
• Packets experience a complex series of – Network hops
– Technologies
– Queuing delays
– …
T1 OC3
test
167 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
84
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Apps See the Network as a “Black Box”, So Does AppTransaction Xpert • For most troubleshooting, complexity can be simplified
• Think of entire network as a single “equivalent” circuit
• Once packet leaves source, the delay can be separated into 3 categories
T1 OC3
– Bandwidth delay
– Latency delay
– Congestion (queuing) delay
• What about Protocol?
– Protocol delay is a function of the sender/receiver
– Packet Loss is handled by the protocol
testBandwidth, Latency,
Congestion
168 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Bandwidth Delay
• Time it takes to stream all bits onto circuit
• This number varies with packet size
• Can be calculated from packet size and circuit bandwidthp
HeadersData
01110010010
2000 bits
2000 bits/second
test…01110010010
What is the bandwidth delay for this packet?
169 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
85
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Calculate Serialization Delay
How long does it take to serialize this packet on to the wire?
Serialization and BW are the same thing
test• If you know the bottleneck bandwidth across a link and the size of each
packet, you know how long it takes to get across the link due to serialization delay (ie: Bandwidth delay)
• Note: we are ignoring latency at this point
170 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Latency Delay
• Time it takes for signal (1st bit) to travel across network– Distance / speed of light
– Normal delays caused by device forwarding
• This is a constant number for all packets
7 ms
1 ms
10 ms
1 ms
10 ms
1 ms1st bit
10 ms
testWhat is the latency for this circuit?What value would “ping” produce?
171 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
86
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Application Turns
• A turn is a change in the flow direction of application data
• Turns magnify circuit latency
• An application with no turns would only experience latency oncepp y p y
• An application with 99 turns would experience latency 100 times
• Simple equation for total latency delay– Latency Delay = circuit latency * (turns +1)
– This equation only works on sequential applications
No Application Turns Many Application Turnstest
pp y pp
172 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Congestion Delay
• Queuing on devices
• Usually caused by other traffic
• Can vary from packet to packety p p
Forwarding rate = 1 packet / ms
packet
test 5 packets in forwarding queue
What is the congestion delay for this device?What is the total delay across this device?
173 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
87
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Determine Queuing/Congestion
Packet should have taken thi tithis time . . .
. . . but ended up taking this time
test
• If a packet should take x time but takes y time, y – x yields queuing/congestion delay
174 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Example Network Delay
• Bandwidth = 1,000 bps
• Latency = 5 seconds
5 seconds
30 seconds
Latency delay
Bandwidth delayClient
T = 0
30,000 bits
10 20 30 40 50
test
15 seconds
Congestion delay
Server50
175 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
88
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Packet Trains
• Stream of packets that are sent together
• Arrive at destination spread out
• Many TCP segments of a larger application messagey g g pp g
• Spreading caused by limited bandwidth of bottleneck link
T = 0 10 20 30 40 50 60
test
10 20 30 40 50 600
176 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Calculating Delay for Packet Trains
• Treat packet trains like 1 big packet
• Same rules apply– First—establish delay caused by bandwidth (stream bits onto wire)
– Second—add latency (last bit of the packet train getting across)
– Third—anything left over is congestion
Application Layer
testTransport Layer
Network Layer
177 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
89
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Calculating Delays—Advanced
• Bandwidth = 1,000 bps
• Latency = 5 seconds
+ 10
+ 5
5
Bandwidth Delay
Latency Delay
Congestion Delay
Tier Processing (Client)
+ 15
+ 5
+ 305
5
ClientT = 0 10 20 30 40 50 60 70 80 90 100
test
Tier Processing (Client)
Tier Processing (Server)
Total Delay = 45+15+20+10+10 = 100
10
10
Server
5,000 bits 10,000 bits8010 20 65 100
178 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Pie Chart
• AppTransaction Xpert creates a pie chart that shows all of the components of delay in one screen
• Shows the total impact on response time due to each individual effect
• Each wedge is the maximum benefit that a single upgrade will provide
Bandwidth
Latency
Congestion
Client
Server
Each wedge is the maximum benefit that a single upgrade will provide
Bandwidth 45
Latency 15
Congestion 20
Client 10
Server 10
test
179 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
90
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Network Effects in App Doctor
• AppTransaction Xpert 16 does not require fully specifying network data for import
• Creates a “Network Effects” pie slice
test
180 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Network Effects
• Specifying Bandwidth will cause AppTransaction Xpert to recalculate the values for the components of delay
test• For more information, see Session 1442: Application Capture and Import Strategies with AppTransaction Xpert
181 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
91
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Calculating Delays
• Two ways to calculate the Pie Chart– Adding up (add the components of delay)
– Squeezing down (reduce effects one at a time)
ClientT = 0 10 20 30 40 50 60 70 80 90 100
test
Server8010 20 65 100
182 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Review of Network Delays
• Different factors can affect network delay– Bandwidth
– Latency
– Congestion
• Some factors are constant and some are variable
• Network delay is not always due to lack of bandwidth
• Important that you understand fundamentals to interpret ACE results
test
183 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
92
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
LAB 1
LAB 1
Calculating
Network
Delaystest
184 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Lab 1: Calculating Network Delays
• Overview– Apply knowledge of sources of delay to a real application
• Objectives– From the raw statistics calculate total bandwidth delay
– From the raw statistics calculate total latency delay
– Compare results with ACE analysis
• Starts on page 64 in your lab manual
test
185 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
93
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Response Time = 26.03 Network Data = 3,199,760 Bytes Size of link = 1,544,000 Kbps
Lab 1: Conclusion
p Bandwidth Delay = (3,199,760*8)/1,544,000 = 16.58 Percentage of Bandwidth = 16.58/26.03 = 63.7%
Turns = 337 Latency = 24.23ms Latency Delay = 24.23*(337+1) = 8.18 seconds P t f L t 8 18/26 03 31 4%
test Percentage of Latency = 8.18/26.03= 31.4%
186 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Calculated that Bandwidth caused 63.7% of response time Calculated that Latency caused 29.8% of response time
Lab 1: Conclusion
test
187 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
94
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Why didn’t the math work out perfectly? Some amount of congestion in trace Some amount of protocol delay in trace
E ti did ’t t f bi di ti l t ffi
Lab 1: Conclusion
Equation didn’t account for bi-directional traffic
Our simple math was very close Important to perform this type of “sanity check”
test
188 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Congestion Delay
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
95
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Congestion Delay
It is somewhat amazing that congestion can be calculatedVery difficult to measure and calculate manuallyACE can measure the effect even without seeing the causeACE can measure the effect even without seeing the causeACE provides 2 important pieces of information for
congestionActual congestion delay for each packet The cumulative effect of congestion on the response time
test
190 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Key Concept
Trace file includes 1 piece of time information Time the packet arrived at the capture probe
ACE includes 2 pieces of time information for each packet Time the packet left the sending side Time the packet arrived at the receiving side
1 piece of time information per packet
test
2 pieces of time information per packet
191 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
96
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Key Concept
ACE knows both the send and receive time Delay analysis requires that both values are known How does ACE derive the second piece of time information?p Trace Merge Single Trace Adjust
test
192 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Trace Merge
Each trace shows what happened locallyOnce clocks are aligned, draw the packets from send to receive
Client
Trace captured at client side
BA C C* D
Trace captured at server
ClientT = 0 2 3 4 5 61 7 8 9 10
A C C*B DT = 0 2 3 4 5 61 7 8 109
test
Server
Trace captured at server side
BA C* D Server0 2 3 4 5 61 7 8 9 10
B DA C*
0 2 3 4 5 61 7 8 109
193 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
97
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Single Side Adjust
With a single trace, ACE only knows what happened locallyMust estimate what happened on network and at other side Based on bandwidth and latency values set during import
ACE ti t h th k t t ACE estimates when the packet was sent ACE estimates when the packet will arrive
Client
Trace captured at client side
ClientT = 0 2 3 4 5 61 7 8 9 10
A B C C* E
T = 0 2 3 4 5 61 7 8 9 10
test
Client
BA C C* E
Server0 2 3 4 5 61 7 8 9 10
194 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Trace Merge
What happens if the import parameters are incorrect Bandwidth is set too low Latency is set too highy g
Might create false packet crossingsThis will confuse analysis
Client
Trace captured at client side
ClientT = 0 2 3 4 5 61 7 8 9 10
A B C D E
T = 0 2 3 4 5 61 7 8 9 10
testBA C D E
Server0 2 3 4 5 61 7 8 9 10
195 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
98
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Key Concept
With BOTH send and receive time ACE knows Actual packet delayOrder of how packets were sent and receivedp
What would happen if we graphed the delays?
testTotal Packet Delay
196 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Packet Size vs. Delay Graph
• Graph the packet size vs. total packet delay
Del
ay (s
ec)
Congestion
Slope is bandwidth*
test
Packet Size (bits)
Pure latency
* Inverse of bandwidth (1/BW)
197 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
99
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Packet Size vs. Delay Graph (cont.)
• Graph Packet Size vs. Total Packet Delay
Del
ay (s
ec)
test
Packet Size (bits)
198 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Viewing Congestion in ACE
• Several good ways to view congestion in ACE– AppDoctor diagnosis
– Summary of Delays
– Congestion drill-down
– Data Exchange Chart
test
199 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
100
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Congestion in the AppDoctor Diagnosis
• Specifically marks category as a Bottleneck
test
200 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Congestion in the Summary of Delays
• Shows the cumulative effect of the congestion on response time
• How much slower was the application because of congestion?
test
201 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
101
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Drilling into Congestion
test• Drilling into congestion can show if the packets are getting queued or
otherwise slowed down
202 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Drilling into Congestion (cont.)
test• Look for “plateaus”
• Always keep in mind the scale of the chart!
• Is delay uniform or random?– Is there any encryption on the link?
203 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
102
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Congestion in the Drill-Down
• Packet Congestion graph
• Retransmission graph
test
204 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Congestion in the DEC
• Network chart layered with Packet Congestion graph
test
205 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
103
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Congestion Review
• Congestion Delay is cause by other traffic on the network
• Always better to merge traces from congested circuits– Congestion values for a trace merge are precise
– Congestion values for a single-side adjust are approximations
• Several ways to see congestion and its effect on the application
• Imagine “squeezing” the congestion out of the trace to calculate its contribution to overall delay
test
206 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Protocol Delay
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
104
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Protocol Delay
• Caused by the artificial throttling of traffic
• Packets are prevented from even getting onto the network
• Caused mainly by TCPy y
test
208 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Why Does TCP Exist?
• TCP offers reliable data transport to applications– Application data is guaranteed to be delivered
– Packets that are “lost” by the network will be retransmitted
• TCP makes applications behave in a network-friendly manner– Throttles the rate at which application sends data
– Ensures that no single user consumes all of the bandwidth
• Both can impact application performance
test
209 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
105
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
What Does Protocol Delay Look Like?• Protocol delay prevents packets from even getting on the network
Application Layer
Transport Layer
Network Layer
Protocol DelayNo Protocol Delay
test
0 2 3 4 5 61 7 0 2 3 4 5 61 7 8 9 10 11 12 13
Protocol Delay
210 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Observe Protocol Behavior
Protocol delay is when the ysending side delays or throttles sending of data (this is different than processing delay)
test• Observe how data leaves in stages indicating Protocol Delay
• Observe other attributes like TCP window size
• Our example here is ridiculously simplified, but real protocol delay is observable in a real trace
211 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
106
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Forms of Protocol Delay
• Critical that you understand mechanisms of TCP
• Several different TCP mechanisms cause can protocol delay– TCP windowing
– Slow Start
– Frozen window
– Nagle’s algorithm
– Retransmissions
– Out of sequence packets
test
212 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Viewing Protocol Delay in ACE
• Several ways to view Protocol Effects –AppDoctor diagnosis
S f D l–Summary of Delays
–Protocol drill-down
–TCP Visualization in DEC
test
213 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
107
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Protocol Delay in the AppDoctor Diagnosis
• Specifically marks category as a Bottleneck
test
214 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Protocol Delay in the Summary of Delays
• Double-click on slice to get more detail
test
215 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
108
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Protocol Delay Drill-Down
• Graphs dynamically appear based on AppDoctor Diagnosis
• Users can manually add or remove graphs
• Protocol Delayy– TCP In-Flight Data
– Retransmissions
– Out of Sequence Packets
– TCP Nagle’s Delay
– TCP Advertised Receive Window
test
216 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Protocol Delay Review
• Protocol delay results from the artificial throttling of traffic
• Several different TCP mechanisms can cause it
• Prevents the packet from even getting onto the network p g g
• Can often be calculated without a trace merge
• Imagine “squeezing” the protocol effects out of the trace to calculate its contribution to overall delay
test
217 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
109
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
LAB 2
LAB 2
Understanding
Components of
Delaytest
218 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Lab 2: Understanding Components of Delay
• Overview– User is transferring files with Microsoft file browser
• Transfer time is very slow
• Want to understand cause of delay
• Objectives– Identify and understand the sources of delay
– Compare slow transfer to fast transfer
• Starts on page 69 in your lab manual
test• Homework– Continue with homework sections of lab manual if you have time
219 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
110
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Lab 2: Conclusion
• File transfer is slow• Primary source of delay is protocol/congestion• Some “potential” problems with TCP but no smoking gun• Look at packet delay graph to evaluate congestion
• Significant congestion on WAN circuit– Base latency is 25ms– Some packets take 125 ms to cross the network
test
220 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Lab 2: Conclusion
• Congestion is primary cause of delay
• Removing other traffic on network fixes problem– Reduces congestion
– Response time decreases
• Might be a case for QoS
Significant Congestion No Congestion
test
221 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
111
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Parallel Effects
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
The “Mysterious” Gray Wedge
• AppTransaction Xpert shows “parallel effects” as a gray pie wedge
test– Transaction has more than one effect taking place at one time
– HTTP is an example of a transaction that exhibits parallelism
223 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
112
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Parallel Effects Drill-Down
• Can show the individual effects with their combined parallel effects together
• Solid bars are the individual effects
• Striped bars are the parallel effectsp p
• Optimizing both components simultaneously will give the benefits of the individual effects AND the parallel effects
T = 0 10 20 30 40 50 60 70 80 90 100
Response Time
Parallel Effects of Bandwidth and Latency
Bandwidth 60%
Latency 10%
Server 20%
Parallel 10%
test
Parallel Effects of Bandwidth and Latency
Parallel Effects of Server and Latency
224 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Understanding Parallel Effects
• Parallel delays may seem counterintuitive at first
• It is a real effect and does happen in the real world
• Real Life Parallel Effects example that you will remember!p y– Reading a paper while eating breakfast
– Cooking a meal (dessert in the oven while steak is on the grill)
• Would be difficult to split these into simple 2 category pie charts– Incorrect to say grilling steak was 50% and dessert was 50% ?
– Suggests that eliminating dessert would cut dinner preparation in half
– Probably more like 25% steak, 25% dessert, 50% both
– Life is parallel
test
Life is parallel
225 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
113
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Types of Applications
• Applications fall into two main categories–Transactional
•St d ff t l ti hi b t k t•Strong cause and effect relationship between packets
•Each application message is dependent on the one before it
•Examples– Database queries
– Web transactions
– File transfers
–Asynchronous
test•No strong cause and effect relation between packets
•Examples– Voice
– Citrix
226 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Types of Applications
• Transactional application can be further divided–Sequential
•Si l ti th•Single execution path
•Application is only doing one thing at a time
–Parallel•Two or more simultaneous execution paths
•Application is doing multiple things at a time
test
227 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
114
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Types of Applications
• Sequential (Adding up method)– Each slice of time can be placed into a single delay category
– The components of delay add up to the total response time
ClientT = 0 10 20 30 40 50 60 70 80 90 100
test
Server
228 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Types of Applications
• Parallel (Squeezing down method)– Some slices of time could be placed in multiple delay categories
– The components of delay would up to more than the total response time
ClientT = 0 10 20 30 40 50 60 70 80 90 100
testServerServer
Connection 1Connection 2Connection 3
229 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
115
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
AppDoctor Individual Effects
Response Time 6 sec
Server 2 sec (33%)
Client 2 sec (33%)
• Summary of Delays for a sequential transaction
( )
Bandwidth 2 sec (33%)
test
230 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
AppDoctor Parallel Effects
• Summary of Delays for a parallel transaction
• Server is now processing while simultaneously sending the large message
test
231 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
116
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
AppDoctor Parallel Effects (cont.)
How does AppDoctor handle multiple effects occurring at the same time?
Response Time
6 sec
Server 2 sec (33%)
test
(33%)
Client 2 sec (33%)
Parallel Effects
2 sec (33%)
The "Parallel Effects" categoryMixed wedge shows total impact due to server processing and bandwidth occurring in parallelBecause there may be many mixed wedges, ACE shows these as gray in the Pie Chart
232 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Understanding Parallel Applications
• Calculating delays for sequential transactions is fairly easy– Each slice of time can be attributed to a single cause
– All the slices add up to the total
• Calculating delays for parallel transactions can be complex– A slice of time may be attributed to multiple sources
– Only the squeezing method works here
• Parallel Effects are subtle and not always easy to visualize
• Parallel Effects do not always occur just because delays overlap
• Aren’t you glad you have AppTransaction Xpert to do this for you
test
233 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
117
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Parallel Effects Example
• Parallel– Two independent connections occur in parallel
– Both must complete for the application to complete
ClientConn. B
A
B B B
testServerServer
Conn. A
234 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Understanding Parallel Effects
• Slide Conventions– Show independent connections as parallel DECs
– Keep in mind that these are happening on top of each other
– Obvious now why you can’t just add delays up
Client
ServerServer
T = 0 10 20 30 40 50 60 70 80 90 100
Connection A
test Client
ServerServer
T = 0 10 20 30 40 50 60 70 80 90 100
Connection B
235 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
118
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Understanding Parallel Effects
• Parallel applications with squeezing method– Perform experiments that squeeze out each category of delay
• Color associated with that delay would disappear
• Response time of transaction may decrease
– Record the difference from each experiment
Client
ServerServer
T = 0 10 20 30 40 50 60 70 80 90 100
test Client
ServerServer
T = 0 10 20 30 40 50 60 70 80 90 100
236 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Understanding Parallel Effects
• What would happen if bandwidth became infinite?– Green would be squeezed out
– Total response time of the transaction would change to 40 seconds
– Bandwidth is responsible for 60 seconds of response time (60%)
Client
ServerServer
T = 0 10 20 30 40 50 60 70 80 90 100
test Client
ServerServer
T = 0 10 20 30 40 50 60 70 80 90 100
237 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
119
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Understanding Parallel Effects
• What would happen if latency became zero?– Red would be squeezed out
– Total response time of the transaction would change to 90 seconds
– Latency is responsible for 10 seconds of response time (10%)
Client
ServerServer
T = 0 10 20 30 40 50 60 70 80 90 100
test Client
ServerServer
T = 0 10 20 30 40 50 60 70 80 90 100
238 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Understanding Parallel Effects
• What would happen if the server became infinitely fast?– Yellow would be squeezed out
– Total response time of the transaction would change to 80 seconds
– The server is responsible for 20 seconds of response time (20%)
Client
ServerServer
T = 0 10 20 30 40 50 60 70 80 90 100
test Client
ServerServer
T = 0 10 20 30 40 50 60 70 80 90 100
239 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
120
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Understanding Parallel Effects
• Something interesting just happened– Bandwidth caused 60%, Latency caused 10%, Server caused 20%
– The individual sources of delay are only responsible for 90%
– What caused the remaining 10%?
• Delays that are not removed by a single source are Parallel Effects
Bandw idth
Bandwidth 60%
Latency 10%
testLatency
Server
Parallel
Latency 10%
Server 20%
Parallel 10%
240 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Understanding Parallel Effects
• Squeeze out 2 effects at the same time– Bandwidth = 60%
– Latency = 10%
– Server = 20%
– Parallel Effects = 10%
Client
ServerServer
T = 0 10 20 30 40 50 60 70 80 90 100
Thread A
test Client
ServerServer
T = 0 10 20 30 40 50 60 70 80 90 100
Thread B
241 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
121
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Understanding Parallel Effects
• Remove the Server and Latency together– Server by itself is 20%, Latency by itself is 10%
– The pair is responsible for 35 seconds of response time (35%)
– Parallel interaction of the 2 provides a bonus benefit if changed simultaneously
Client
ServerServer
T = 0 10 20 30 40 50 60 70 80 90 100
Thread A
test Client
ServerServer
T = 0 10 20 30 40 50 60 70 80 90 100
Thread B
242 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Understanding Parallel Effects
• Remove Bandwidth and Latency together– Bandwidth by itself is 60%, Latency by itself is 10%
– The pair is responsible for 75 seconds of response time (75%)
– Parallel interaction of the 2 provides a bonus benefit if changed simultaneously
Client
ServerServer
T = 0 10 20 30 40 50 60 70 80 90 100
Thread A
test Client
ServerServer
T = 0 10 20 30 40 50 60 70 80 90 100
Thread B
243 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
122
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Parallel Effects Drill-Down
• Can show the individual effects with their combined parallel effects together
• Solid bars are the individual effects
• Striped bars are the parallel effects
• Optimizing both components simultaneously will give the benefits of the individual effects
Bandwidth 60%
Latency 10%
Server 20%
Optimizing both components simultaneously will give the benefits of the individual effects AND the parallel effects
Response Time
T = 0 10 20 30 40 50 60 70 80 90 100
Parallel Effects of Bandwidth and Latency
testParallel 10%
Parallel Effects of Server and Latency
244 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Parallel Effects Summary
• Delays that can not be removed by a simple change of one parameter
• To reduce, you must change at least two parameters simultaneously
• These delays are not bad y– Application is efficient
– Application is doing multiple things at the same time
testBandw idth
Latency
Server
Parallel
245 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
123
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
User Think Time
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Capturing a Good Transaction (Task)
• Important to get an “atomic” transaction– Atomic transaction is the smallest transaction that you can capture effectively
– Capture should be from edge of user input to screen refresh
– Eliminate “PEBCAK”
test
247 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
124
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Atomic Transaction Example
• Modify Customer Record– Login
• Call up the Login screen• End user complains• Type information
• “Submit”
– Look up record• Call up correct screen/menu
• Type information
• “Submit”
– Modify data• Mouse/cursor around
End user complains “modifying a customer record is slow”
• Need to find out what is involved in transaction
test• Type information
• “Submit”
– Logout• Call up Logout screen
• “Submit”
Red arrows indicate the key part of the transaction to capture
248 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Identifying and Removing User Think Time
• Always get as close as you can to a perfect “Atomic” transaction as described in the Capture Section p
• However, sometimes you just can’t get the perfect capture
• “User think time” enters the trace
• AppTransaction Xpert deals with think time– Auto detects when calling up AppDoctor
– Allows user to explicitly specify think time
test
249 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
125
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Removing User Think Time
AppTransaction Xpert automatically identifies that the trace contains user think time. This dialog comes up when AppDoctor is selected
AppTransaction Xpert
testalso allows the user to pick any dependency on the client (first talker) and categorize it as user think time
250 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Removing User Think Time
Original Analysis
testUser Think Time Removed
251 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
126
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
LAB 3
LAB 3
Understanding
Parallel Effects
test
252 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Lab 3: Understanding Parallel Effects
• Objectives– Explore an application transaction that has significant parallel effects
• Trace is fairly simple
• Trace is somewhat artificial
– Practice visualizing the squeezing method to understand the behavior
– Use Drill-down and QuickPredict to understand the interactions
• Starts on page 84 in your lab manual
test
253 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
127
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Lab 3: Conclusion
• Parallel delays may seem counterintuitive at first
• It is a real effect and does happen in the real world
• This example is exaggerated to make it easy to visualizep gg y
test
254 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Agenda Overview of TCP
- TCP Basics
- TCP Windowing
- Lab 4: Examining TCP Behavior
T bl h ti TCP Troubleshooting TCP
- Multi-User QuickPredict
- Bandwidth-Delay Product
- Lab 5: Troubleshooting a Slow File Transfer
- Detecting Packet Loss
- Frozen Window
- Nagle’s Algorithmtest
- Nagle s Algorithm
Discrete Event Simulation
Instructor Demo: Nagle’s Algorithm using DES
References
- What’s New ?
- Wrap up255 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
128
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
TCP Basics
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
What is TCP?
• Transmission Control Protocol
• Transport layer protocol p y p
• Layered between IP and application layer
• Typical applications such as
– HTTP (Web)
– POP, IMAP, SMTP (e-mail)
Application Application
TCP
IP
test– Telnet, FTP, and more
Data Link
257 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
129
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Why Does TCP Exist?
1) TCP offers reliable data transport to applications
• User data is guaranteed to be delivered
2) TCP makes applications behave in a network-friendly manner
• TCP may throttle the rate at which it sends user data
Both impact application performance
test
258 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Which Applications Don’t Use TCP?
• TCP adds overhead
• TCP limits throughput
• Not all applications need what TCP providespp p– VoIP – Voice over IP
– Video Conferencing
– SNMP – Simple Network Management Protocol
– Network Gaming Protocols
– Domain Name System (DNS)
– Network Time Protocol (NTP)
test
259 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
130
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Motivation – Why Is TCP Important?
Example TCP/IP Stack
Application Developers HTTPSOAP
HTTPSOAP
IPFIBER
IPETHERNET
IPETHERNETNetwork Managers
TCP TCP
Application Developers
SSLHTTP
SSLHTTP
test• TCP is in a key location of networked applications’ protocol stacks
– Frequently misunderstood– This makes people and software that understand TCP valuable
260 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Review: ACE Application Message Chart
Timeline
Application message groups
Application message arrows
Tiers
test
Red packets have no payload (usually acknowledgements, SYNs, or FINs)
Packet sizes
g p
261 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
131
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
How to Read TCP Decodes
Size is size of entire packet (including IP, MAC, etc.)
Receive window
Special flagsSequence number
Destination port
Receive window
F b t k ftest
Acknowledgement number
Payload length
Source portFrame number taken from trace file(s). May not always be in strict order due to trace merge
262 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Opening a Connection
• Three-way handshake to open connection
• Used to exchange– Initial SEQ numbers
Cli– TCP Options Client
1A
2
3C
test
Server
2B
263 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
132
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Closing a Connection
• Four-way handshake to close connection• “I have no more data to send.”• Typical example
client273B
274C
connection closed at client
test
server
272A
B C
275D
connection closed at server
264 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Closing a Connection (cont.)
• Often only three segments used
• Second FIN also is an acknowledgement for first FIN (B&C in one packet)
• TCP Resets are analternate way toterminate a connection 86
A
87B&C
88D
test
B&C
265 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
133
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
SEQ and ACK Numbers
• Ensures that data was received
• Sending TCP gives each segment a sequence number– SEQ of segment K is SEQ of segment K-1 plus length of segment K-1
• Receiving TCP acknowledges reception by replying with ACK– ACK number is the SEQ number of the next expected byte
• SEQ and ACK numbers are inbytes, not packets
Receiver
SEQ=100 LEN=10
Sender
test
Q
ACK= ?
SEQ = ? LEN=55
110
110
K-1 K
Seq # K-1 Seq # K
266 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
SEQ and ACK Quiz
• Which frame(s) is 36 ACKing?
• Which frame(s) is 44 ACKing?
test
267 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
134
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
TCP Acknowledgements
• Receiver normally ACKs every other packet– Inefficient to ACK every packet
• What if only one packet is received?– Delayed-ACK timer must expire
– 100-200 ms (OS determines exact delay)
– Usually does not impact total application performance
emen
t
test
Dat
a
Ack
now
ledg
e268 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Review: TCP Basics
• TCP is a connection-based protocol
• It uses a three-way handshake for opening connections
• Each side closes the connection separatelyp y
• SEQ and ACK numbers ensure reliable delivery
test
269 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
135
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
TCP Windowing
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
TCP Windowing
• TCP window size is the maximum amount of data TCP will allow outstanding (“in-flight”)
• If window size is less than or equal to outstanding data, TCP will not send data
• In trace below, server is waiting for ACKs before sending more data
Client
test
Server
271 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
136
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
TCP Windowing
• TCP uses the minimum size of the three windows– Congestion control window
• Changes during connection
• Internal to sending TCP
• Not visible in real world
• Lives in the “mind” of the sender
– Receive buffer
• May change during connection
• User-configurable
• Visible as WIN= in decodes
test– Send buffer
• Controlled by sending
application
• Not visible in decodes
• Rarely the bottleneck
Congestion Control Window
Receive Buffer
In-Flight Data
272 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
TCP Congestion Control Window and Slow Start
• Estimate of how much data the network can handle
• Sending TCP increases the congestion control window each time it receives g gan ACK– Congestion control window increases exponentially every round-trip time
• Actually stored in bytes: 1460, 2920, 5840, 11680
• When TCP first starts transmitting data, it goes slowly– Sender starts by sending one segment, then waits for ACK
test
273 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
137
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
TCP Congestion Control Window in Slow Start
CWND = 1Data (S1)
Sender Receiver
CWND = 2
CWND = 3
CWND 4
ACK (S2)
Data (S3)Data (S2)
Data (S5)Data (S4)
Data (S6)
ACK (S4)
testCWND = 4
CWND = 7
ACK (S8)
Data (S7)Data (S8)Data (S9)
ACK (S10)Data (S10)
ACK (S12)
Data (S11)Data (S10)
274 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Bulk Data Transfer
• Congestion window is greater than receive window
• TCP sends data as soon as it receives acknowledgements
• In-flight data is constantg
• In-flight data = outstanding data = unacknowledged bytes
• Data transfer limited by receive window
test
When you see a pattern like this, blame TCP275 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
138
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Review: TCP Windowing
• TCP limits the amount of outstanding data
• Minimum of three windows is used– Send window: How much data the sender can handle
– Receive window: How much data the receiver can handle
– Congestion control window: How much data the network can handle
• Congestion window– Initially small (1, 2, or 4 segments)
– Grows exponentially during slow start
test
276 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
LAB 4
LAB 4
Examining
TCP Behavior
test
277 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
139
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Lab 4: Examining TCP Behavior
• Use AppTransaction Xpert to examine TCP behavior
• You will see– Connection opens
– Slow start
– Congestion control window
• Raise your hand, and a TA will assist you with any questions
• Starts on page 95 in your lab manual
test
278 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Lab 4: Conclusion
• TCP uses SEQ and ACK numbers to ensure reliable delivery
• Minimum of three windows (congestion, receive, and send) determines whether TCP sends more data
• Comparison of bulk and chatty applications– Window size and bandwidth determine throughput for bulk applications
– Latency determines throughput for chatty applications
test
279 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
140
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
When to Suspect TCP as a Bottleneck
• You should suspect TCP of causing problems when:– You cannot explain why an application is running slowly
– The servers and network are adequately provisioned
• Servers have low load
• Sufficient bandwidth
• Links not congested
– The ACE Live™ Response Time Composition Chart shows
• Large Payload Transfer (Xfer) Times
• Large Retransmission Delays
– AppTransaction Xpert ’s AppDoctor Summary of Delays tab shows
test• Large Network Transfer Effects
• Large Protocol delays
– AppTransaction Xpert ’s AppDoctor Diagnoses point to
• One or more TCP-related bottlenecks or potential bottlenecks
280 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Troubleshooting TCP with ACE Live- Excellent for focusing troubleshooting, monitoring TCP health
• Where to start looking, where and when to capture• Measures connection resets, packet loss, long response times, and more
ACE Live Response Time Composition Chart
– Large Retransmission DelayLarge Retransmission Delay
• May be sign of packet loss
• May also be unfiltered duplicate packets
– Large Payload Transfer Times
• May be TCP-related
• May also indicate bandwidth, congestion or even server processing causes
test
281 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
141
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Troubleshooting TCP with AppTransaction Xpert
• AppDoctor provides quick way to check for TCP problems
• Suspect TCP if the Summary of Delays shows– Large Network Transfer Effects
– Large Protocol delays
• Use diagnosis tab to identify whichspecific TCP-related issue affectsyour trace– Bottlenecks
– Potential bottlenecks
test
282 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Understanding Network Transfer Effects
• Combination of three effects related to transferring data– Bandwidth, Congestion, Protocol
• Bottleneck bandwidth is needed to separate these phenomena– Refine Network Effects key to identification of delays due to the TCP protocol
test
283 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
142
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
How to Identify TCP Problems
• Look at the AppDoctor Diagnosis tab– Effect of Protocol
– Effect of Network Transfer
– Last 7 stats deal with TCP
– Helps target the specificTCP issue
• Look at decodes for long
application transactions
• Look in the networktest
• Look in the network
chart at the start and
end of long delays
284 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
How to Identify TCP Problems (cont.)
• Double-click in the Summary of Delays for more information
• Use separate drilldown screens for Protocol and Congestion
test
285 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
143
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
How to Identify TCP Problems (cont.)
• Congested networks also can cause protocol delays• TCP implements congestion avoidance• Single-sided captures of receiving data may make congestion look like
protocol delay or tier processing
test
286 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Review: Analyzing TCP Behavior
• Look for TCP problems in traces with large amount of protocol delay
• Congestion delays also induce protocol delays
• AppDoctor (forensic)pp ( )– Separately diagnoses protocol and congestion
– Diagnoses several TCP problems
• QuickPredict (what if?)– Separately models protocol and congestion
– Can simulate changing TCP Window Size
test
287 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
144
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Review: ACE QuickPredict• Quick simulation engine
– Combination of analytical and discrete event simulation– Uses analytical models of TCP behavior
• Initially shows a pristine view of network• Allows you to study effects of the network on the applicationAllows you to study effects of the network on the application• Outputs graphs of response time vs. one of the following:
– Bandwidth– Latency– Packet loss– Link utilization– TCP receive window size– NEW: WAN Acceleration parameters
test• QuickPredict Bar Charts supports
asymmetric window sizes
288 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Review: QuickPredict Input and Output
• Y axis is always response time
• X axis is a range of values for one of:– Bandwidth, Latency, Packet Loss, Link Utilization, TCP Window size,
and WAN Acceleration (NEW!)
Change values of other variablesto see impact on response time
test
289 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
145
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Multi-User QuickPredict
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Multi-User QuickPredict (MUQP)
• Single-user QuickPredict considers one client only
• MUQP allows you to have more than one– Client and client site
– Applications (either ACE traces or ACE Whiteboard traces)
– Data centers
• Output– Response time (including
– delay components)
– Utilization
• Results
test
Results– Average
– N-th percentile
291 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
146
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Inputs to MUQP
• Deployed applications (defined in ACE and/or ACE Whiteboard files)
• Network topology (number of client sites, clouds, and data centers)
• Tier-pair parameters (bandwidth, latency, link utilization, window size, and p p ( ypacket loss)
• Application deployment options (number of transactions per hour for each client site)
test
292 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Outputs of MUQP
• "Average-case" response times for applications
• "Worst-case" results for access links (utilization) and applications (total response time and components of tier processing, latency, protocol, and congestion delay)
test
293 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
147
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Instructor Demo of MUQP
test
294 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Bandwidth-Delay Product
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
148
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Bandwidth-Delay Product• Bandwidth-delay product — Defined as amount of data (in bytes) that can be
sent before ACK is received for the first unacknowledged bit of data – 2 x one-way latency x bandwidth of slowest link
• To “fill the pipe,” TCP window size must be equal to bandwidth-delay product• If window is too small, TCP must wait for acknowledgment before sending new datag g• Example below
– Bandwidth: 1,544 kbps = 193 KB/sec– Latency: 250 ms = 0.250 sec– BWDP: 2 x .25 x 193 = 96.5 KB– Receive Window: 8 KB (1/12 ideal size)– Throughput limited to: 130 kbps (1/12 link speed)
test
296 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Bandwidth Delay Product = 2 × Capacity = 24 KB (16 segments)
Capacity = .096 s × 125 KB/s = 12 KB (8 segments)
Bandwidth-Delay Product
Sender Receiver
Bandwidth-Delay Product = 2 × Capacity = 24 KB (16 segments)
96 ms
test
1 Mbps
297 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
149
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
LAB 5
LAB 5
Troubleshooting
a Slow File Transfer
test
298 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Lab 5: Troubleshooting a Slow File Transfer
• Overview – Computer manufacturer moving bulk data from Chicago to Taipei has dedicated
T1 without any packet drops
– Not seeing expected response time
– You are a network engineer
– Your manager: “It must be the network”
– Contemplating (expensive) link upgrade
• Objective– Determine the cause of slow
response time
test– Propose and test a solution
• Starts on page 107 in your lab manual
299 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
150
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Lab 5: Conclusion
• Simple TCP configuration change improved performance
• Over same time period:p– Large receive window
test– Small receive window
• What is the downside of making this change?
300 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Changing the Receive Window
• How do I change the receive window on my computer?– Windows
• http://www.microsoft.com/technet/network/tcpip/default.mspx
• http://www.microsoft.com/technet/itsolutions/network/deploy/depovg/tcpip2k.mspx
• Google search: windows tcp/ip implementation details
– Solaris
• http://www.sean.de/Solaris/soltune.html#water
– Linux
• http://www.speedguide.net/read_articles.php?id=121
test• Default for modern operating systems is 64 KB
– Maximum advertised receive window value in TCP packet is 64 KB
– How do we handle today’s large capacity, high latency pipes?
• TCP Window Scaling
301 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
151
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
TCP Window Scaling
• TCP allows for WIN= field to be scaled by a power of 2– Decodes will show WIN<<2=262140– WIN<<x means multiply field by 2x
• This setting is exchanged in TCP options at beginning of connectionThis setting is exchanged in TCP options at beginning of connection– Both sides must agree and understand this option
test
• What if a packet trace does not contain the opening of a connection?– AppTransaction Xpert cannot detect the presence of TCP Window scaling
302 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Detecting Packet Loss
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
152
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
TCP and Packet Drops
• Where do packets get dropped, and why?
– Media collisions
– Wireless signal interference
– Congested networks
• Endpoints must react to overloaded routers
– Drops are a form of indirect communication between routers and senders
• Routers have multiple strategies for keeping queue size in check
– Tail drop
– Random Early Detection (RED)
test
304 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Packet Loss and Retransmissions
Packet is dropped by the network and retransmitted Time between when it should have arrived and when it did arrive Controlled by TCP timers
C t d P t l D l Counted as Protocol Delay Retransmission also invokes congestion control Slows down throughput Counted as Protocol Delay
ClientT = 0 2 3 4 5 61 7 8 9 10
A B B
test
Server0 2 3 4 5 61 7 8 9 10
Protocol Delay
305 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
153
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Drops and Retransmissions
• TCP must make sure all bytes of a stream are received– TCP guarantees delivery by retransmitting packets presumed dropped
• Two methods by which TCP will retransmit a dropped packetTwo methods by which TCP will retransmit a dropped packet
–Retransmission timeout• Sending TCP schedules a timer for outstanding packets
• Timer is cancelled when the packet is acknowledged
• If timer expires, packet is presumed missing, re-sent
–Triple duplicate ACKs• ACK number indicates next expected packet
test• If receiving TCP does not get the expected packet, it ACKs every packet
• If receiving TCP does not get the expected packet, ACK number does not increase.
• If sending TCP receives acknowledgements with the same ACK number, it knows a packet was likely lost
306 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
TCP One Loss: Retransmission TimeoutSender Receiver
Data (S104)
Data (S103)Data (S102)Data (S101)
ACK (S103)Data (S104)
ACK (S105)
Data (S108)
Data (S107)
Data (S106)
Data (S105)
( )
ACK (S106)
test
Data (S106)
ACK (S109)
Retransmission timeout
ACK (S106)
307 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
154
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
TCP One Loss: Triple Duplicate ACK
Sender Receiver
Data (S103)Data (S102)Data (S101)
ACK (S103)
Triple duplicate ACK and fast
Because 104 was not received, Receiver does not change ACK number
Data (S104)
ACK (S104)
Data (S107)
Data (S106)
Data (S105)
ACK (S103)
ACK (S104)
ACK (S104)
Data (S108)
test
ACK and fast retransmit
Data (S104)
ACK (S109)
( )
ACK (S104)
308 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Cost of Loss Detection
• Retransmission timeout– There are not enough ACKs for triple duplicate ACK– Cost: Timer (often large) plus time to send 1 packet– Congestion window decreases to one segment
• Triple-duplicate ACKs– Cause fast retransmit– Cost: Time to send 1 packet
Triple-duplicate ACK
Retransmission
test
• Notice the send times Retransmission
309 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
155
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Review: Detecting Packet Loss
• Two mechanisms– Retransmission time-out (RTO)
• Costly
• Always works
– Triple-duplicate ACK
• Relatively efficient
• Does not always happen
• Chatty applications rarely “qualify”
• Food for thought– Why acknowledge every packet during loss?
test– Why not quintuple-duplicate ACK?
– Why not single-duplicate ACK?
310 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Frozen Window
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
156
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
TCP Frozen Window
• Receiver stores packets in receive buffer
• Advertised receive window “WIN=” reduced by data in buffer
• Normal applications remove data quicklypp q y
• What happens if:– A server is handling too many clients?
– What if the server is running other applications?
– What if a another thread is hogging the CPU?
test
312 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
TCP Frozen Window (cont.)
• As WIN decreases, buffer fills up, sender cannot send data
• Connection becomes “frozen”
• Sender will send probe packets to see if window size has increasedp p
test
313 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
157
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
TCP (Almost) Frozen Windows
• As WIN decreases, sender cannot send data
• What if the WIN is less than MSS?
• Sender could send data, but often does not
• Detection of “chilly windows” new to AppTransaction Xpert 16.0
test
314 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Drilling into Protocol (cont.)
test
• Frozen Windows:– Receiving WIN goes to zero– Sending side sends “probe” packets
315 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
158
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Nagle’s Algorithm
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Nagle’s Algorithm
• Nagle’s Algorithm– TCP mechanism to reduce sending of many tiny messages
– Sender may not send data unless a full, MSS-sized segment can be sentOR there is no unacknowledged data
• Delayed Acknowledgement
• Interaction of these two mechanisms sometimes causes problems
• Ridiculous situation where both sides wait for the other to do somethingClient
testB
Server
A B
200 ms317 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
159
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Nagle’s Algorithm Example
• Client sends segment A (less than MSS)
• Client wants to send segment B, but needs to wait for ACK
• Server is waiting for another segment before sending ACKg g g
Eventually, server sends ACKClient then sends segment BProcess repeats…
Client
testB
Server
A B
200 ms318 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Nagle’s Algorithm Example (cont.)
• What causes the delay for this application?
test
319 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
160
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Nagle’s Algorithm and Delayed ACK
• To eliminate delays when sending consecutive small messages:– Rewrite the application
• Sending multiple small consecutive messages is inefficient
• Sometimes you can’t change the application
– Disable Nagle’s algorithm
• Done within the application
• Effects only the application changed: (TCP_NODELAY socket option)
• Not always possible to change an existing application
– Acknowledge every packet
• Practical real world alternative
test• On Windows: http://smallvoid.com/article/winnt-nagle-algorithm.html
• Warning: Applies to every application on the machine!– Interacts with congestion control window, affecting bulk applications
320 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
References
• TCP in OPNET– Online documentation– Overview PDF
• Model Library/Standard / TCP Model Usage GuideACE TCP Wi d i B ttl k li d t ti– ACE TCP Windowing Bottleneck online documentation
• Applications in OPNET– Several whitepapers including SMARTE
• Help / Product Documentation / Modeling Methodologies / Methodologies and Case Studies
• http://www.psc.edu/networking/projects/tcptune/– Setting system wide TCP window sizes– Listing of TCP features for each operating system
• http://www.ncsa.uiuc.edu/People/vwelch/net_perf/tcp_windows.htmlInstructions on setting TCP window sizes within application code
test
– Instructions on setting TCP window sizes within application code• http://www.sean.de/Solaris/soltune.html
– Tuning your TCP/IP stack on Solaris• TCP/IP Illustrated, Volume I, Richard Stevens
321 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
161
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Take-Away Points
• Windowing causes TCP to wait before transmitting application data
• Bulk data-transfer throughput often limited by TCP windowing
• TCP frequently does not use the full available bandwidthq y
• Packet loss can cause significant reduction in throughput
• QuickPredict is a fast, easy, accurate way to predict application response time
• Use DES to model and study detailed behavior
• Simulate using the proper TCP machine settings
Next step: Run homework labs for a greater understanding of TCP
test
322 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
What’s New ??
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
162
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
What’s New in AppTransaction Xpert ??
• 64-Bit Support– Access more that 4GB of memory on 64-bit machines which allows you analyze
much larger capture files
• Windows 7 Professional (32-bit and 64-bit)
• Capture Enhancements– Manage captures from ACE Live appliances within the Application Capture
Manager. New ACE Live tab to quickly download and preview data
– ACE Capture Agent 3.9 Ability to filter the traffic data by time range, IP addresses, and/or IP conversations.
– Capture File Filters Greater flexibility in specifying hosts and ports.
test• Support for Riverbed Transparent Mode
– The AppTransaction Xpert import now supports capture files from Riverbed accelerators running in full or port transparency mode
324 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
What’s New in AppTransaction Xpert (cont.)
• Streamlined Import Process– Quickly open capture files through the user interface without specifying any information
(such as location, bandwidth, latency, etc.)
• Bandwidth Latency and Locations can be specified as neededBandwidth, Latency, and Locations can be specified as needed
– Automatic Filtering of Duplicate Packets and TSO (TCP Segmentation Offloading)
– Simplified Manual Merge Workflow
• QuickPredict Enhancements - WAN Acceleration Support– Support for WAN optimization technologies by allowing users to predict behavior for
compression, caching, and application acceleration
• HTTP Anal sis Enhancementstest
• HTTP Analysis Enhancements– Automatic detection of HTTP on non-standard ports (thus eliminating the need to perform
redecodes or modifying config files)
– Improved analysis for packet slicing and missing packets
– HTTP-specific visualization of object response times in TreeView
325 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
163
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
References
test
OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Additional Documentation• Online at:
www.OPNET.com / Support / Support Center
–Documentation
–Release Notes
• Also look at: http://userforums.opnet.com
–Info on ALL OPNET products
–Discussion Groups
Hi & i ktest
–Hints & tricks
327 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
164
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
OPNET Training Courses
•OPNET provides Instructor-led training for all of our products. –OPNET Training Facilities;OPNET Training Facilities;
• Bethesda, Cary, Dallas, Nashua, Santa Clara
–On-site at YOUR facilities–On the Web via WebEx
•For AppTransaction Xpert :–2-day Instructor-led-training
test•Register for courses online at: www.opnet.com/Training
328 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
OPNET Certification Program
Become recognized as an expert user of OPNET products…
Certification covers relevant topics such as:
• Management / AdministrationManagement / Administration
• Configuration
• Product Selection
• Product Operation
• Networking Knowledge
Upon successful completion:
test
p p
• Letter
• Certificate
• Individual certification ID
• Valid for 2 years329 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
165
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
AppTransaction Xpert Online Training
test
330 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
OPNET Web Pages - Training Resource
test
331 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
166
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
OPNETWORK is the largest event of its kind, attracting thought leaders from industry, government, and academic communities from all over the world to
test
332
RSVP at: www.opnet.com/opnetwork2010
and academic communities from all over the world, to advance best practices for:
• Application Performance Management• Network Engineering, Operations, and Planning• Network R&D
332 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
Course Evaluation
We’ll use this information to improve our courses and teaching methods. Please enter as much comment material as you’d like… the more info you add, the better we’ll get!
Point your browser to:
http://eval.opnet.com
Please enter Instructor and Course info…– Course: Introduction to AppTransaction Xpert – Course Date:– Instructor Name:
FREE
Stuff !!test
Thank You in advance for your participation!
333 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
2/10/2011
167
OPNET Confidential – Not for release to third parties © 2010 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.
test
334 OPNET Confidential – Not for release to third parties © 2011 OPNET Technologies, Inc. All rights reserved. OPNET and OPNET product names are trademarks of OPNET Technologies, Inc.