IBM FlashSystem – An Application Accelerator By Doug Herman – hermand@us.ibm.com

What benefits will IBM FlashSystem give to a high performance application?

Lower Latency: I/O response time of 25-100 micro-second range

Processor Efficiency: Reduce I/O wait, increase processor efficiency

Enhance end user experience: Improve application response

Application Types that can benefit from IBM FlashSystem:

OLTP Databases

Analytical Application

HPC/Computational Applications

Cloud-scale Infrastructures

IBM FlashSystem is last year’s acquisition of Texas Memory Systems. They have 34 years

of experience in low latency SSD/Flash industry. The latest products are the 19th

generation

of flash technology. This 1U appliance can have 20 TB usable capacity delivering 450K

IOPS.

Reliability, Availability & Serviceability

eMLC Flash has 10x more endurance than MLC flash

Patented VSR (Variable Stripe Reliability) allows continued operation, in the event

of a Flash DIMM failure

2D Raid configuration to minimize service disruptions by 10x

33% over-provisioned capacity – gives the industry highest availability

How is IBM Flash Storage connected?

First, this does not disturb the existing storage infrastructure, meaning that it is fully

compatible with IBM Storage, EMC, and Hitachi products. Second, it will complement by

offloading the heavy I/O requirements from the existing storage infrastructure.

The 1U appliance can be connected into existing SAN infrastructure using fibre channel

connectivity. Flash LUNs are configured and zoned to the Logical Partition (LPAR) direct

attached or through Virtual I/O server using NPIV. The LPAR running AIX will use LVM

mirroring.

Flash storage systems

IBM FlashSystem 710 and 810

IBM FlashSystem 720 and 820 for High Availability

For more details: http://www-03.ibm.com/systems/storage/flash/

POWER SYSTEMS NEWSLETTER

UNITED STATES - EAST

June 2013

Send questions, comments, or concerns: richard.milton@us.ibm.com (or to subscribe)

In this issue:

IBM FlashSystem (pg1)

The Math behind Server

Virtualization (pg 1)

PowerHA (pg 3)

Useful URLs (pg 3)

A New Choice for VIOS

Migrations and Upgrades (pg 3)

IBM PowerLinux for Big Data

Analytics (pg 5)

PowerSC Real-Time

Compliance (pg 6)

3rd Party Endorsements

Selecting a Strategic Operating

System

Success Stories

Business Impacts of SAP

Deployments

Value of Virtualization

Future of the Datacenter

Community

flickr

Twitter

LinkedIn

YouTube

Facebook

RSS Feed

The Math behind Server Virtualization

By Dave Levites – dlevites@us.ibm.com

Is it the “Virtualization of POWER”…or the power of virtualization that provides the real performance benefits of deploying multiple

workloads on a single Power Server? Well, I’m here to tell you: it’s both. The more workloads that run on a server, then the

greater the likelihood that one VM’s peak usage will align with another VM’s valley. There is math (statistical multiplexing) behind

this reasoning to help establish the true value. In addition to that, there are costs for delivering that capability, which I call the

virtualization tax, that have to be added to the tab.

(Continued on page 2)

In a nutshell, the math works this way. Determine the average (“M”, for mean) and

peak (“P”) utilization of a server. These numbers can then be used to determine the

standard deviation for that workload. There’s a rule in statistics (“normal distribution”,

aka “68, 95, 99.7” rule) that basically states what percent of potential values would be

within 1, 2, or 3 times the standard deviation away from the mean. That formula is

Peak minus the Mean, divided by the number of deviations (n), divided by the Mean,

or: “Sn=(P-M) / n / M “.

So, for example, if the mean was 5% busy and the peak was 37%, the “2-Sigma”

would be ((37%-5%)/2)/5%, or 3.2. A server would have to provide the equivalent of

21% of current capacity -- 5% + (3.2 * 5%) -- to meet the workload demand, 95% of

the time.

Now, there’s another rule (called the Central Limit Theorem) that states when you

pool “n” normal distributions, the total standard deviation of the additions only

increases by the square root of “n”.

(“The Math behind Virtualization” … continued from page 1)

So, for instance, if there are 4 workloads sized as above, the 95th

percentile peak required to run all of them on the same server

would not be 4 * 21%...it would be 52% or “5% * 4 + (3.2 * 5%) * √4”. Likewise, 8 workloads would require about 85%. Extending

that a bit, further yields:

Means Peak Sigma-2 Workloads 95th

%tile 5% 37% 3.2 1 21% “ “ “ 4 52% “ “ “ 9 93% “ “ “ 16 144% “ “ “ 25 205%

Of course, the cost of virtualization needs to be added to these numbers, and that’s where the “Virtualization of POWER” comes in

play. First off, Power servers scale to extreme sizes (up to 256 cores), so by their very nature of being large, they greatly reduce

the amount of CPU needed to support a particular workload. For example, if we made an assumption that the 8 workloads above

consumed 85% of an 8-core server, say a p720, then a p795 driven up to a 90% utilization level could accommodate over 500 such

instances. That’s the equivalent of over sixty 8-way servers.

In addition to scalability, there’s the matter of virtualization performance. Several studies (Edison Group & Solitaire Interglobal

come to mind) have shown a lower overhead associated with PowerVM versus any of the software virtualization methods deployed

on Intel x86 servers. An example of this is one such test that was performed internally by IBM’s Software Group a year or so ago

where they ran a Websphere workload on a competing x86 server and on a Power 750. They produced a chart which showed a

significant virtualization overhead penalty – nearly 10% of the server horsepower while running just 4 VMs. The Power server, on

the other hand delivered near linear performance as the workload scaled from 1 to 4 partitions.

Performance, utility, TCO are all important value dimensions of virtualization...and the math behind it can help you to add it all up.

Page 2

Useful URLs:

Prerequisite Site - https://www-912.ibm.com/e_dir/eserverprereq.nsf

IBM Systems Energy Estimator - http://www-912.ibm.com/see/EnergyEstimator

AIX Whitepapers - http://www-03.ibm.com/systems/power/software/aix/whitepapers/index.html

AIX Release Strategy - http://www-03.ibm.com/systems/power/software/aix/support/release_strategy.html

Power Systems Technical Guide - http://www-03.ibm.com/systems/power/hardware/reports/factsfeatures.html

AIX Movies - https://www.ibm.com/developerworks/mydeveloperworks/wikis/home?lang=en#/wiki/Power%20Systems/page/Hands-

On%20Technical%20Movies

Power Code Matrix - http://www-304.ibm.com/support/customercare/sas/f/power5cm/home.html

PowerHA: Providing Availability for your Infrastructure Annie Broadhurst - ambroad@us.ibm.com

Today we live in a world driven by the continuous need for data. With so many smart devices available to consumers, businesses

are feeling additional pressure to provide information at a non-stop basis. Creating the likelihood of taking downtime, whether it is

planned or unplanned, in the infrastructure environment has become increasingly difficult. This is how PowerHA comes into the

picture for Power System environments.

PowerHA is a high availability and disaster recovery solution, providing clients with the ability to eliminate

a single point of failure. It insures that the client data is not permanently unavailable to the end user during

planned and unplanned outages. This can be done during the failover process when data is transferred from

one node to another node in a cluster, allowing for near continuous availability.

There is a wealth of information available online to help you plan, develop, and/or maintain a highly available

Power Systems environment.

PowerHA Homepage Video Demonstration Redbooks

A New Choice for VIOS Migrations and Upgrades

By Bob Foster - bobf@us.ibm.com

Many clients worry about, and thus avoid upgrades. Several IBM clients using Virtual I/O Server (VIOS) will be interested a new

offering called Simplified VIOS Migration/Upgrade to help them simply and easily move to an updated and supported version of

VIOS.

Some History

IBM PowerVM technology has been adopted and used successfully worldwide by IBM clients. One of the major building blocks of

this technology is VIOS, which allows virtualization of client workloads. Once clients have virtualized their workloads, they are

offered a huge range of functions and features they can use for managing and controlling their workloads and server utilization.

Since VIOS helps to exploit so much of the PowerVM technologies, it’s a mainstay of most PowerVM environments. IBM’s first

release of VIOS was back in 2004 with Version 1.1. Since then, IBM has been adding features to VIOS and shipped its last Version

1 (Version 1.5) in November 2007, and its most recent version 2.2 in 2013. New functions and features are exclusively added to

newer versions of the VIOS and older versions are unable to use the new features. IBM also has a limited time that it will offer

support for each level of VIOS, so older levels are unable to get fixes if a problem is found. For example, support for VIOS 1.5

ended in 2011 and support for VIOS 2.1 ended in 2012.

Clients may have multiple reasons why they may not have kept their VIOS upgraded. One of the primary reasons is the lack of a

maintenance window during which an upgrade can be performed. Additionally, if they are still on VIOS 1.X, they probably would

want to upgrade to a supported level of VIOS 2.2 as upgrading to VIOS 2.1 still leaves them running an unsupported level of VIOS.

This causes another problem as upgrading to a more recent level from much older levels has more steps involved in the process.

If clients are still at VIOS 1.5, they are now two major releases behind and running levels of VIOS that shipped five years ago.

Some of the reasons they may not have been at current levels have compounded themselves. Now, the maintenance window will

be much longer for the upgrade and the process itself will require multiple steps to get to the newest supported levels.

(Continued on page 4) Page 3

(“A New Choice for VIOS Migrations and Upgrades” … continued from page 3)

New Offering Delivers Savings

IBM’s VIOS development team and the IBM Lab Services team have developed the Simplified VIOS Migration/Upgrade offering,

which includes both a process and a toolset that will make a VIOS upgrade much simpler and quicker.

Here’s an example of the required procedure for upgrading from VIOS 1.5.1 to VIOS without the toolset:

1. Update VIOS 1.5.1 to VIOS 1.5.2.1-FP-11.12.2.1.5 2. Apply the Service Pack to update VIOS 1.5.2.1-FP-11.1 SP-02 3. Migrate the VIOS to VIOS version 2.1.3 by using the Migration DVD 4. After the VIOS is at version 2.1.3, apply Fix Pack 2.2.1.1 5. Apply the Service Release to bring the VIOS to level 2.2.1.5

Instead, consider this Simplified VIOS Migration/Upgrade offering procedure for the same upgrade to VIOS 2.2.1.5:

1. Run scripts on the VIOS 1.5.1 to capture mapping and ODM data

2. Install the VIOS with a mksysb of VIOS 2.2.1.5

3. Run post-install scripts to restore the mapping and ODM data

Not only does the offering eliminate steps, it greatly simplifies the tasks required through the use of scripts. This process can be

also be used to upgrade to any newer levels of the VIOS.

The toolset is customized for each client based on the specific environment. The first factor is the level of VIOS to be migrated from

and the final VIOS level desired by the client. The second factor is that clients can (and normally do) have different levels of VIOS

on their servers and the tool used to capture the vscsi mappings needs to be tested on those different levels of VIOS. Finally, the

client’s SAN storage and multipathing software (MPIO) levels being used in the VIOS are also an important factor. There are device

settings that are required for the different MPIO software that need to be rebuilt on the updated VIOS. These three factors create a

large number of possible combinations, so IBM develops and tests the toolset per engagement.

Three Phases of Engagement

There are three phases of this engagement. First, the client supplies IBM with environment data. As mentioned, this includes

current levels of VIOS to upgrade and the final VIOS level, types of SAN storage and the levels of MPIO software in VIOSs.

The second phase begins when, with this information, IBM development and Lab Services recreate the client’s environment (as

exactly as possible) and build the toolset. The toolset will include a version of viosbr (a utility that ships in newer VIOS levels but

didn’t exist in older levels and is used to back up and restore settings) that can be used on the older versions of VIOS to capture

the VIOS vscsi and virtual Ethernet mappings. Most SAN environments will require a script that can pull the appropriate device

attributes and set those values after the VIOS is updated (post-install script). Depending on the combination of SAN/MPIO/VIOS

levels, this phase could take a few weeks. If the combination is similar to a previous client’s environment, this phase could be

reduced.

The third phase involves Lab Services and the client working together at the client site to build the appropriate process for the

environment. The client and IBM will create a VIOS mksysb including the new levels of MPIO/SAN software required. Using this

mksysb and the toolset, the client and IBM test the process and adjust accordingly. Once the toolset and process are verified, the

client and Lab Services will perform a live upgrade.

Stay Up-To-Date

Clients need to keep their PowerVM environment up-to-date and at supported levels of VIOS, so any issues that may occur can be

handled by support and have fixes supplied if needed. IBM can now help simplify that upgrade and require a much shorter

maintenance window with this new offering.

Are you still running AIX 5.3?

AIX 5.3 TL12, SP8 now supports POWER7+ servers. Customers must have a valid AIX Software Maintenance Agreement with a

Support Extension for AIX 5.3 contract. See announcement letter 213-146 for additional details.

Models affected:

IBM Power 710 (8231-E1D)

IBM Power 720 (8202-E4D)

IBM Power 730 (8231-E2D)

IBM Power 740 (8205-E6D)

IBM Power 750 (8408-E8D)

IBM Power 760 (9109-RMD)

NOTE: Power 770 (9117-MMD) & Power 780 (9179-MHD) were already supported by AIX 5.3 TL12, SP7. Page 4

Why PowerLinux for Big Data Analytics? By Kevin McCombs – mccombsk@us.ibm.com

There is no question that Big Data Analytics is exploding in today's world. Just read the papers. Regardless of

what your opinion is concerning the recent news that the government may be using data from many sources to

identify potential terrorist activities, one has to agree that the ability to use all of this disparate data for analytics is mind boggling.

Imagine the scope if all cell phone activity, data usage, internet usage, email, etc were to be archived and mined. What kind of

performance would you need to crunch data of that magnitude in a timely matter? We are talking zetabytes of data! Now that’s

Big Data, but I digress.

Big Data Analytics can basically be broken into two categories, Data in motion and data at rest. Data in motion includes constantly

changing streaming data sources that require real time or low latency analytics. The PowerLinux solution for that would be IBM

InfoSphere Streams. Data at rest would include many diverse static datasets. IBM's Watson is a great example of a large

PowerLinux cluster that uses data at rest. Watson used Open Source Apache Hadoop for this solution. IBM also offers IBM

InfoSphere BigInsights as an enterprise ready, out-of-the-box, Hadoop based solution.

Why would one chose PowerLinux over Linux on x86 for this task? In a nutshell, it’s POWER7+. POWER7+ offers double the

number of threads/core than x86's (4 vs 2). POWER7+ has a 45% higher clock speed that x86 (Sandy Bridge). Power 7+ has 4

times more L3 cache than x86. POWER7+ provides the high throughput memory and I/O bandwidth required for Big Data. IBM

also offers a JVM tuned specifically for POWER7+. Lastly, the least known advantage to most customers, POWER7+ has a list

price LOWER than x86.

Did you know that PowerLinux can sort 1TB of data in less than half the time of its x86 competition? In 2012 a 10-node Hadoop

cluster of IBM PowerLinux 7R2 servers was able sort through a terabyte of data in less than 9 minutes. In 2012 we reran the

benchmark with the new Power7+ 7R2. A 10 node cluster (16 core/node) reduced that number to under 6.7 minutes, a sorting rate

of 1.04 GB/min/core. In contrast, an 18 node Cloudera Hadoop cluster of HP ProLiant Gen8 DL380 could only muster a rate of

0.57 GB/min/core in a recent benchmark:

(http://www.hp.com/hpinfo/newsroom/press_kits/2012/HPDiscover2012/Hadoop_Appliance_Fact_Sheet.pdf ).

PowerLinux is a strong emerging technology for Big Data and a more than worthy opponent for Linux on x86 in many diverse

workloads. Learn more today at:

http://www-03.ibm.com/systems/power/software/linux/

(IBM PowerLinux 7R2)

Tools for Planning and Documentation

HMC Scanner - https://www.ibm.com/developerworks/community/wikis/home?lang=en#!/wiki/Power+Systems/page/HMC+Scanner

NMON Analyser - https://www.ibm.com/developerworks/community/wikis/home/wiki/Power%20Systems/page/nmon_analyser

IBM Systems Workload Estimator - http://www-947.ibm.com/systems/support/tools/estimator/index.html

IBM System Planning Tool - http://www-947.ibm.com/systems/support/tools/systemplanningtool/

IBM Energy Estimator - http://www-912.ibm.com/see/EnergyEstimator

IBM Stencils - http://www.visiocafe.com/ibm.htm Page 5

PowerSC Real Time Compliance By Stephen Dominguez – sdoming@us.ibm.com

The November 2012 PowerSC release has provided an extremely powerful and sophisticated

monitoring solution for AIX systems called Real Time Compliance, RTC.

As its name suggests, RTC provides immediate notification of security events related to a set of

monitored files. The default install of RTC monitors a set of 278 standard AIX files that are critical to

system security. This set of files can be changed on the fly without a reboot or need to refresh the

RTC daemon. You can even monitor important files installed on your AIX partition related to 3rd

party

products.

On a per file basis, RTC can monitor for up to two types of security events for a particular file. One

type of security event is a content change event. The other type of event is an access change event.

The following is an example of a content change event:

Pretend a hacker gains access to a system and edits, using vi, sensitive security files such as

/etc/security/user and /etc/security/login.cfg. As soon as each of these files is changed, an email

message will immediately be sent to the RTC administrative email address configured for that

system.

The message will actually detail what user changed the file, by “userID” and “groupID”. If the user

changed userIDs using the su command, the email message will even detail what was the original

login ID of the user that edited the file. The message will also detail that vi was used to change the

content of the file.

The following is an example of an access change event:

Pretend a hacker gains access to a system and changes the file permissions, using chmod, of

sensitive security files such as /etc/inetd.conf and /etc/inittab. As soon as each of these files is

changed, an email message will immediately be sent to the RTC administrative email address

configured for that system. The email will contain the same type of detail as described previously for

the content change event.

The real magic of RTC is gained when you also implement PowerSC’s Security and Compliance

Automation feature. RTC actual works in cooperation with the Security and Compliance Automation

feature to provide security monitoring capabilities never before possible.

This is how it works:

Pretend you deploy the HIPAA profile of PowerSC’s Security and Compliance Automation feature in

order to help your AIX partitions to be more HIPAA compliant.

Deploying that profile will change the policy of the default minimum length of passwords to 8

characters. Let’s also suppose you have deployed RTC. Now if a hacker gains access to your

system and goes into /etc/security/user and changes the “minlen” attribute from 8 to 0, not only will

RTC send you a notification that /etc/security/user has changed, but because you have also

deployed the Security and Compliance Automation feature, when RTC notices a content change in

/etc/security/user, RTC will trigger the aixpert engine to re-check the security policies configured by

the Security and Compliance Automation feature. When all of this is completed, not only will you get

an email about the file content change, but aixpert will use its capabilities and identify that the HIPAA

policy for the default minimum length of passwords has been changed and the email will detail that

the actual value of 8 has been changed to zero.

In Summary, Real Time Compliance is a truly powerful AIX security monitoring feature. To learn

more about the PowerSC consulting services available to assist you in deploying and integrating

PowerSC please contact me, Stephen Dominguez, at smsdoming@us.ibm.com

POWER7+ Server Redbooks (aka “D” models)

Power 710 and 730 Technical Overview and Introduction

Power 720 and 740 Technical Overview and Introduction

Power 750 and 760 Technical Overview and Introduction

Power 770 and 780 Technical Overview and Introduction

Page 6