Date post: | 28-Jan-2015 |
Category: |
Technology |
Upload: | cloudyn |
View: | 106 times |
Download: | 0 times |
More Bang for your Buck
Sharon Wagner Co-‐Founder & CEO, Cloudyn Microso: Cloud Day April 23rd 2013
How to get the most out of your cloud
Who am I?
• Director, Products, Oblicore • Sr. principal, Cloud Business Unit, CA • Co-‐Founder & CEO, Cloudyn™
Cloud Economics
@cloudyn_buzz
Cloudyn.com | blog.cloudyn.com
How do you measure your cloud efficiency?
Overwhelming informaOon
# of deployment cycles, idle servers, costs, % u9liza9on, jobs, applica9on performance, speed, databases, network bandwidth, server performance, end user performance, applica9on response 9me, revenue, green benefits.
Cool. Now we have a good grasp on
our cloud!
Usage trend: compute uOlizaOon
Most virtual servers are significantly underutilized.
• Average yearly CPU uHlizaHon of 17% • Max RAM uHlizaHon of 64%
By looking at CPU, Memory, I/O, Network:
Usage trend: pricing
Most customers use the least economical pricing modules.
• 26% reserve capacity upfront • 71% runs on-‐demand virtual instances
How customers pay cloud vendors:
93% of the on-‐demand instances should be reserved.
Actual usage, perfect and over provisioning
How do you measure your car efficiency?
Keep it simple
The only thing that maSers is the business impact!
User Experience =
Workload
Capacity
OK, lets keep it really simple
?
Houston, we have a problem
Cloud capacity is
∞
Cloud Value =
Cost
Compute Units
The right metric
What does this mean? What should I do?
1. Think small – rightsizing 2. Go green – recycle and reHre 3. Use your cash – reserve capacity
Three steps to get more out of your cloud
Nego?a?on has zero impact…
How can we find opOmizaOon opportuniOes?
Bringing real cloud usage data from 500 IaaS customers into the mix:
~2.5m Virtual instances, thousands of databases and billions of storage objects monitored in the survey.
Recycling sample: storage/compute
Surprise! You have storage. • Typically represents 15% of cloud spend • Only 12% are using cheaper storage opHons • 15% of volumes are disconnected
Does it make sense to keep the ligh
t
on when you leave the roo
m?
Sizing sample
One large cappuccino with an extra 5GB of RAM please… Coffee customizaHon, Starbucks @ IaaS
If you do it to your coffee, why not treat your instances the same? It’s 20% of your monthly cost.
Sizing sample: compute
Comparing large to x-large for RDBMS workload:
Spec Large X-‐Large RAM 7.5 GB 15 GB CPU 4 EC2 CU 8 EC2 CU Storage 850 GB 1690 GB I/O Performance Moderate High
Large instance, volume-‐opHmized with 500 MBPS provisioned IOPS performed beSer than single
x-‐large and cost less!
Pricing opOmizaOon
*Cloud vendors love charging less
* Yep, this is not a typo, but they have to run a business too… So, how does it work?
Pricing trend: reserved, on-‐demand
You will have to reserve your capacity
93% of the on-‐demand instances should be reserved.
• Requires one Hme payment, pay less for each hour • Resource availability is guaranteed • 71% of instances runs on-‐demand, 26% runs reserved
Price opOmizaOon
Why they love charging you less? • Because they have no choice • Customer saHsfacHon • The upfront payment
• Capacity projecHon • The Jevons paradox Hi, It’s me, William
Stanley Jevons
$-‐
$200
$400
$600
$800
$1,000
$1,200
$1,400
$1,600
$1,800
$2,000
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov
VM Cost
Time
On-‐Demand Reserve X Month Reserve Y Month
Large instance: on-‐demand vs. reserved capacity comparison Savings
Breakeven aoer 4mon, 30% RunHme
Common reservaOon mistake
Safe ReservaHon
OpHmal ReservaHon
Summary: the cycle