SlideShare a Scribd company logo
© 2014 Amazon.com, Inc. and its affiliates. All rights reserved. May not be copied, modified, or distributed in whole or in partwithout the express consent of Amazon.com, Inc. 
November 12, 2014 | Las Vegas 
ENT203Iterating Your Way to 95% Reserved Instance Usage 
Toban Zolman, Cloudability
Today’s topics 
•Introduce a Reserved Instancepurchasing methodology 
•Understand how RIs are applied 
•Learn the best metrics to measure RI effectiveness 
•How to adjust reservations and infrastructure to achieve 95% coverage 
Toban Zolman 
VP, Product DevelopmentCloudability 
toban@cloudability.com
@cloudability 
Cloud cost analytics 
Cloudability 
1 
2 
3 
4 
5 
Spend 
management 
Spending visibility via monthly forecasts, dash- board, budget alerts, and daily email digests. 
Cost 
analytics 
Dig into your operating costs with detailed costs by tag, service, and usage type. 
EC2 usage 
analytics 
Spot underutilized resources w/ instance level usage metrics. 
RI purchase 
analytics 
Understand the exact combination of reserved instances that will maximize your savings. 
Enterprise 
enablement 
Organizational group views/filtering/rollups, multi-user access. 
$1B+of cloud spending under management • 13,000 users
RI buying methodology
Why RIs? And why 95% coverage? 
•RIs can be the easiest way to control costsAs a Product & Engineering head,I’d much rather use RIs to reduce cost than work on other efficiencies 
•Every dollar spent on on-demand could be 50% cheaper with an RIWith the exception of unplanned or short-lived usage you should be running almost everything under a reservation 
•With scalable architecture its not practical to have full 100% coverageA monthly buying and modification cycle can net you at about 95% coverage over the course of a year
What drives infrastructure changes? 
•Application needsApps change, scale, and become more sophisticated needing more compute, memory,or storage 
•Price changesAs Amazon drives down prices,it changes the cost profile of existing infrastructure 
•New optimized instancetypesLeverage optimized instances to be more efficient 
The goal is to anticipate these items and constantly tweak reservations to maximize coverage (and thus savings)
Bet the house on reactionary purchases
too littletoo late
BURNING MONEY
How is your company structuring RI buys?
Reservation management methodology
Reservations 
Infrastructure
Buy 
Measure 
Learn 
Purchase 
Align 
ROI 
Minimize time 
through loop
Buy
Topics 
•Understanding reservations 
•Purchasing strategies
Understanding reservations
Cost savings 
Capacity reservation 
Reservations have two parts
Reservations are applied each hour
What makes up a reservation? 
Instance 
RI term 
RI type 
Instance family 
Size 
OS 
Location (AZ) 
m1 
large 
Linux 
us-east-1a 
1 year 
Light 
r3 
8xlarge 
RHEL 
us-west-2b 
3 years 
Medium 
m3 
medium 
Windows 
ap-northeast-1b 
1 year 
Heavy 
Source: http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/reserved-instances-fundamentals.html
Each combination has a different price 
There are 2,000+ different reservation configurations 
•Each with their own up-front fee 
•Each with their own hourly usage price 
•Each with their own “break-even” points 
Source: http://aws.amazon.com/ec2/pricing/
Purchasing strategies
Buy early and often
Fear, uncertainty, doubt stalls decisions 
•Should I wait until prices drop? 
•What if our usage changes? 
•What if I don’t want to be in a contract for three years?
Actual AWS price history 
$0.00 
$0.25 
$0.50 
$0.75 
$1.00 
3/1/09 
12/1/09 
9/1/10 
6/1/11 
3/1/12 
12/1/12 
9/1/13 
6/1/14 
1 Year RI 
3 Year RI 
On Demand
Actual AWS price history (effective rate) 
$0.00 
$0.25 
$0.50 
$0.75 
$1.00 
3/1/09 
12/1/09 
9/1/10 
6/1/11 
3/1/12 
12/1/12 
9/1/13 
6/1/14 
1 Year RI 
3 Year RI 
On Demand
Total price 3 years of use 
•Prices drop, but not fast enough to erode saving 
•If usage is greater than 50% 3 year RIs saved money 
Source: http://blog.cloudability.com/cloud-cost-war-shouldnt-stop-buying-reserved-instances/
Analyze worst case scenario 
What if we bought an RI the day before a price change? 
Old price 
New price 
Change 
On demand 
$0.26 
$0.208 
20% 
1 year ri 
$0.064 
$0.056 
12.5% 
February 2013 price change (m1.large medium RI 80% usage): 
Total cost 
1 Year RI before change 
$1,201 
On demand after change 
$1,458 
New on-demand price still higher than old RI price 
Total cost before and after 
Source: http://blog.cloudability.com/cloud-cost-war-shouldnt-stop-buying-reserved-instances/
Cost of waiting for a change 
Extra cost for 4 months of OD 
Savings from locking in new price 
Amount over paid 
$387 
$154 
$233 
Source: http://blog.cloudability.com/cloud-cost-war-shouldnt-stop-buying-reserved-instances/
It doesn’t pay to wait 
CONCLUSION
Scenario 2: Purchase frequency 
0 
13 
25 
38 
50 
1 
3 
5 
7 
9 
11 
13 
15 
17 
19 
21 
23
On-demand 
Annual 
Quarterly 
Monthly 
Total cost 
$309,053 
$205,209 
$134,900 
$95,854 
Discounted hours 
0% 
48% 
90% 
95% 
Savings 
0% 
34% 
63% 
69% 
$40k savings 
Scenario 2: Purchase frequency 
Source: http://blog.cloudability.com/why-you-need-to-buy-aws-reservations-more-than-once-year/
Purchase RIs as frequently as possible 
CONCLUSION
Buy 
Measure 
Learn 
Minimize time 
through loop
How to structure monthly purchases 
1.Cover as many hours as you can as quickly as you canThe faster you can get coverage the more you will save over time 
2.Focus your budget on high confidence purchasesBuy reservations for instances that are (almost) always running 
3.Focus on the highest savings firstBuy reservations for instances that have a rapid ROI
Identify instances with high savings and heavy utilization
How to structure monthly purchases 
1.Cover as many hours as you can as quickly as you canThe faster you can get coverage the more you will save over time 
2.Focus your budget on high confidence purchasesBuy reservations for instances that are (almost) always running 
3.Focus on the highest savings firstBuy reservations for instances that have a rapid ROI 
4.Spread purchases horizontally across infrastructurePurchase tactically across all instance families to avoid cliffs and lock-in
Measure
Instance coverage
(ENT203) Iterating Your Way To 95% Reserved Instance Usage | AWS re:Invent 2014
Focus measurement on two factors: Instance coverage &reducing hourly cost
Learn
Migrate, modify, or mature
Migrate
Migration is about building an RI friendly architecture: 
Increase flexibility &maximize cost savings
1. Consolidate instance families
Why instance familiesmatter 
•RI modifications can only happen within a familyReservations can be changed at any time from one size to another within a family 
•RIs can apply across consolidated billing accountsReservations can apply across linked accounts if the original account is not using it. Fewer families means more spots an RI can be applied.
2. Migrate to modern instances
Cost savings 
m1.medium 
m3.medium 
On demand 
$0.087 
$0.07 
Heavy RI upfront 
$247 
$222 
Heavy RI hourly 
$0.02 
$0.018 
Annual cost (100% util) 
$762 
$613 
Save $150yr 
20%savings 
Source: http://aws.amazon.com/ec2/pricing/
Instance type migration paths 
Legacy instance type 
Modern instance type 
T1 
T2 
M1 
M1 
M3 
C1 
C3 
H1 
I2 
M2 
R3 
CR1 
Source: http://aws.amazon.com/ec2/previous-generation/
Migration strategy 
•Proactively plan instance type migrationsFinance and Ops have to talk on this one to understand when upgrades should happen and what impact it has on reservations 
•Tactically renew reservations on legacy instancesIf you’re going to run a legacy instance for a while,you might as well cover it with an RI but only renew what you have to 
High frequency reservation purchases make this transition easier
M1 reservations can still be a good deal 
Buy RIs on legacy instance types ifyou are still going to be using them in 6 months (or shop in the marketplace for shorter term RIs) 
savings
Modify
Modifying Reserved Instances 
•Instances with Linux OSInstance type (within family) Availability Zone (within region) Network (VPC or Classic) 
•Instances with a licensed OS or without a familyAvailability Zone (within region) Network (VPC or Classic) 
•Reservations cannot be moved between accounts
We arefamily 
M1* 
M2* 
M3 
C1* 
C3 
R3 
i2 
t2 
smallmediumlargexlarge 
xlarge2xlarge4xlarge 
mediumlargexlarge2xlarge 
mediumxlarge 
largexlarge2xlarge4xlarge8xlarge 
largexlarge2xlarge4xlarge8xlarge 
xlarge2xlarge4xlarge8xlarge 
microsmallmedium 
* LEGACY FAMILY
Instance types without a family 
•t1.micro* 
•cr1.8xlarge* 
•cc2.8xlarge* 
•cg1.4xlarge* 
•cr1.8xlarge* 
•hi1.4xlarge* 
•hs1.8xlarge 
•g2.2xlarge 
* LEGACY FAMILY
Modifying instancetype 
Instance size 
Normalization factor 
micro 
0.5 
small 
1 
medium 
2 
large 
4 
xlarge 
8 
2xlarge 
16 
4xlarge 
32 
8xlarge 
64
2xlarge 
Modifying instance type 
=16PTS
2xlarge 
xlarge 
xlarge 
Modifying instance type 
16 
8 
8 
+ 
=
2xlarge 
xlarge 
xlarge 
large 
large 
large 
large 
Modifying instance type
medium 
medium 
medium 
medium 
medium 
medium 
medium 
medium 
2xlarge 
xlarge 
xlarge 
large 
large 
large 
large 
Modifying instance type
small 
small 
small 
small 
small 
small 
small 
small 
small 
small 
small 
small 
small 
small 
small 
small 
medium 
medium 
medium 
medium 
medium 
medium 
medium 
medium 
2xlarge 
xlarge 
xlarge 
large 
large 
large 
large 
Modifying instance type
small 
small 
small 
small 
small 
small 
small 
small 
small 
small 
small 
small 
small 
small 
small 
small 
medium 
medium 
medium 
medium 
medium 
medium 
medium 
medium 
2xlarge 
xlarge 
xlarge 
large 
large 
large 
large 
Modifying instance type
Reservations represent sunk cost 
Modifications let you move those reservations tomaximize savings as your infrastructure changes
(ENT203) Iterating Your Way To 95% Reserved Instance Usage | AWS re:Invent 2014
Amazon EC2 Reserved Instance Marketplace 
If you simply can’t use existing reservations then selling them on the marketplace is a viable option
Mature
Coverage is king
Conclusion
Buy 
Measure 
Learn 
Purchase 
Align 
ROI
Take aways 
•Buy RIs early and often 
•Focus budget on “high confidence” purchases 
•Measure RI coverage and cost-per-hour metrics 
•Migrate, modify,and mature reservations to achieve 95% coverage
http://bit.ly/awsevals

More Related Content

(ENT203) Iterating Your Way To 95% Reserved Instance Usage | AWS re:Invent 2014

  • 1. © 2014 Amazon.com, Inc. and its affiliates. All rights reserved. May not be copied, modified, or distributed in whole or in partwithout the express consent of Amazon.com, Inc. November 12, 2014 | Las Vegas ENT203Iterating Your Way to 95% Reserved Instance Usage Toban Zolman, Cloudability
  • 2. Today’s topics •Introduce a Reserved Instancepurchasing methodology •Understand how RIs are applied •Learn the best metrics to measure RI effectiveness •How to adjust reservations and infrastructure to achieve 95% coverage Toban Zolman VP, Product DevelopmentCloudability toban@cloudability.com
  • 3. @cloudability Cloud cost analytics Cloudability 1 2 3 4 5 Spend management Spending visibility via monthly forecasts, dash- board, budget alerts, and daily email digests. Cost analytics Dig into your operating costs with detailed costs by tag, service, and usage type. EC2 usage analytics Spot underutilized resources w/ instance level usage metrics. RI purchase analytics Understand the exact combination of reserved instances that will maximize your savings. Enterprise enablement Organizational group views/filtering/rollups, multi-user access. $1B+of cloud spending under management • 13,000 users
  • 5. Why RIs? And why 95% coverage? •RIs can be the easiest way to control costsAs a Product & Engineering head,I’d much rather use RIs to reduce cost than work on other efficiencies •Every dollar spent on on-demand could be 50% cheaper with an RIWith the exception of unplanned or short-lived usage you should be running almost everything under a reservation •With scalable architecture its not practical to have full 100% coverageA monthly buying and modification cycle can net you at about 95% coverage over the course of a year
  • 6. What drives infrastructure changes? •Application needsApps change, scale, and become more sophisticated needing more compute, memory,or storage •Price changesAs Amazon drives down prices,it changes the cost profile of existing infrastructure •New optimized instancetypesLeverage optimized instances to be more efficient The goal is to anticipate these items and constantly tweak reservations to maximize coverage (and thus savings)
  • 7. Bet the house on reactionary purchases
  • 10. How is your company structuring RI buys?
  • 13. Buy Measure Learn Purchase Align ROI Minimize time through loop
  • 14. Buy
  • 15. Topics •Understanding reservations •Purchasing strategies
  • 17. Cost savings Capacity reservation Reservations have two parts
  • 19. What makes up a reservation? Instance RI term RI type Instance family Size OS Location (AZ) m1 large Linux us-east-1a 1 year Light r3 8xlarge RHEL us-west-2b 3 years Medium m3 medium Windows ap-northeast-1b 1 year Heavy Source: http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/reserved-instances-fundamentals.html
  • 20. Each combination has a different price There are 2,000+ different reservation configurations •Each with their own up-front fee •Each with their own hourly usage price •Each with their own “break-even” points Source: http://aws.amazon.com/ec2/pricing/
  • 22. Buy early and often
  • 23. Fear, uncertainty, doubt stalls decisions •Should I wait until prices drop? •What if our usage changes? •What if I don’t want to be in a contract for three years?
  • 24. Actual AWS price history $0.00 $0.25 $0.50 $0.75 $1.00 3/1/09 12/1/09 9/1/10 6/1/11 3/1/12 12/1/12 9/1/13 6/1/14 1 Year RI 3 Year RI On Demand
  • 25. Actual AWS price history (effective rate) $0.00 $0.25 $0.50 $0.75 $1.00 3/1/09 12/1/09 9/1/10 6/1/11 3/1/12 12/1/12 9/1/13 6/1/14 1 Year RI 3 Year RI On Demand
  • 26. Total price 3 years of use •Prices drop, but not fast enough to erode saving •If usage is greater than 50% 3 year RIs saved money Source: http://blog.cloudability.com/cloud-cost-war-shouldnt-stop-buying-reserved-instances/
  • 27. Analyze worst case scenario What if we bought an RI the day before a price change? Old price New price Change On demand $0.26 $0.208 20% 1 year ri $0.064 $0.056 12.5% February 2013 price change (m1.large medium RI 80% usage): Total cost 1 Year RI before change $1,201 On demand after change $1,458 New on-demand price still higher than old RI price Total cost before and after Source: http://blog.cloudability.com/cloud-cost-war-shouldnt-stop-buying-reserved-instances/
  • 28. Cost of waiting for a change Extra cost for 4 months of OD Savings from locking in new price Amount over paid $387 $154 $233 Source: http://blog.cloudability.com/cloud-cost-war-shouldnt-stop-buying-reserved-instances/
  • 29. It doesn’t pay to wait CONCLUSION
  • 30. Scenario 2: Purchase frequency 0 13 25 38 50 1 3 5 7 9 11 13 15 17 19 21 23
  • 31. On-demand Annual Quarterly Monthly Total cost $309,053 $205,209 $134,900 $95,854 Discounted hours 0% 48% 90% 95% Savings 0% 34% 63% 69% $40k savings Scenario 2: Purchase frequency Source: http://blog.cloudability.com/why-you-need-to-buy-aws-reservations-more-than-once-year/
  • 32. Purchase RIs as frequently as possible CONCLUSION
  • 33. Buy Measure Learn Minimize time through loop
  • 34. How to structure monthly purchases 1.Cover as many hours as you can as quickly as you canThe faster you can get coverage the more you will save over time 2.Focus your budget on high confidence purchasesBuy reservations for instances that are (almost) always running 3.Focus on the highest savings firstBuy reservations for instances that have a rapid ROI
  • 35. Identify instances with high savings and heavy utilization
  • 36. How to structure monthly purchases 1.Cover as many hours as you can as quickly as you canThe faster you can get coverage the more you will save over time 2.Focus your budget on high confidence purchasesBuy reservations for instances that are (almost) always running 3.Focus on the highest savings firstBuy reservations for instances that have a rapid ROI 4.Spread purchases horizontally across infrastructurePurchase tactically across all instance families to avoid cliffs and lock-in
  • 40. Focus measurement on two factors: Instance coverage &reducing hourly cost
  • 41. Learn
  • 44. Migration is about building an RI friendly architecture: Increase flexibility &maximize cost savings
  • 46. Why instance familiesmatter •RI modifications can only happen within a familyReservations can be changed at any time from one size to another within a family •RIs can apply across consolidated billing accountsReservations can apply across linked accounts if the original account is not using it. Fewer families means more spots an RI can be applied.
  • 47. 2. Migrate to modern instances
  • 48. Cost savings m1.medium m3.medium On demand $0.087 $0.07 Heavy RI upfront $247 $222 Heavy RI hourly $0.02 $0.018 Annual cost (100% util) $762 $613 Save $150yr 20%savings Source: http://aws.amazon.com/ec2/pricing/
  • 49. Instance type migration paths Legacy instance type Modern instance type T1 T2 M1 M1 M3 C1 C3 H1 I2 M2 R3 CR1 Source: http://aws.amazon.com/ec2/previous-generation/
  • 50. Migration strategy •Proactively plan instance type migrationsFinance and Ops have to talk on this one to understand when upgrades should happen and what impact it has on reservations •Tactically renew reservations on legacy instancesIf you’re going to run a legacy instance for a while,you might as well cover it with an RI but only renew what you have to High frequency reservation purchases make this transition easier
  • 51. M1 reservations can still be a good deal Buy RIs on legacy instance types ifyou are still going to be using them in 6 months (or shop in the marketplace for shorter term RIs) savings
  • 53. Modifying Reserved Instances •Instances with Linux OSInstance type (within family) Availability Zone (within region) Network (VPC or Classic) •Instances with a licensed OS or without a familyAvailability Zone (within region) Network (VPC or Classic) •Reservations cannot be moved between accounts
  • 54. We arefamily M1* M2* M3 C1* C3 R3 i2 t2 smallmediumlargexlarge xlarge2xlarge4xlarge mediumlargexlarge2xlarge mediumxlarge largexlarge2xlarge4xlarge8xlarge largexlarge2xlarge4xlarge8xlarge xlarge2xlarge4xlarge8xlarge microsmallmedium * LEGACY FAMILY
  • 55. Instance types without a family •t1.micro* •cr1.8xlarge* •cc2.8xlarge* •cg1.4xlarge* •cr1.8xlarge* •hi1.4xlarge* •hs1.8xlarge •g2.2xlarge * LEGACY FAMILY
  • 56. Modifying instancetype Instance size Normalization factor micro 0.5 small 1 medium 2 large 4 xlarge 8 2xlarge 16 4xlarge 32 8xlarge 64
  • 58. 2xlarge xlarge xlarge Modifying instance type 16 8 8 + =
  • 59. 2xlarge xlarge xlarge large large large large Modifying instance type
  • 60. medium medium medium medium medium medium medium medium 2xlarge xlarge xlarge large large large large Modifying instance type
  • 61. small small small small small small small small small small small small small small small small medium medium medium medium medium medium medium medium 2xlarge xlarge xlarge large large large large Modifying instance type
  • 62. small small small small small small small small small small small small small small small small medium medium medium medium medium medium medium medium 2xlarge xlarge xlarge large large large large Modifying instance type
  • 63. Reservations represent sunk cost Modifications let you move those reservations tomaximize savings as your infrastructure changes
  • 65. Amazon EC2 Reserved Instance Marketplace If you simply can’t use existing reservations then selling them on the marketplace is a viable option
  • 69. Buy Measure Learn Purchase Align ROI
  • 70. Take aways •Buy RIs early and often •Focus budget on “high confidence” purchases •Measure RI coverage and cost-per-hour metrics •Migrate, modify,and mature reservations to achieve 95% coverage