AWS re:Invent 2016: Getting the most Bang for your buck with #EC2 #Winning (CMP202)

© 2016, Amazon Web Services, Inc. or its Affiliates. All rights reserved.

Joshua Burgin, General Manager, EC2 Spot, Amazon Web Services

November 2016

CMP 202

Get the Most Bang for Your Buck

with #EC2 #Winning

Amazon EC2 purchasing options

On-Demand

Pay for compute capacity by the

hour with no long-term

commitments

For spiky workloads, or to define

needs

Reserved

Make a 1 or 3 Year commitment

and receive a significant

discount over On-Demand

For committed or baseline

utilization

Spot

Pay market price for unused

compute capacity at a steep

discount over On-Demand

For fault tolerant, time-insensitive or

transient workloads

Pillars of performance and cost-optimization

Right sizing Purchasing

options

Increase

elasticity

Measure,

monitor, &

improve

Right sizing

Right sizing

• Selecting the cheapest instance available

while meeting performance requirements

• Looks at CPU, RAM, storage, and network

utilization to identify potential instances that

can be downsized

Increase elasticity

Turn off non-production instances

• Look for dev/test, non-prod instances that are

running always-on and turn off

Automatically scale production

• Use Auto Scaling to scale in and out based on

demand and usage (for example, spikes)

Measure, monitor, and improve: Uncover the

cost-optimization opportunities

Auto-tag resources

Identify always-on non-prod

Identify instances to downsize

Recommend Reserved Instances to purchase

Dashboard our status

Report on savings

AWS pricing principles

Pay as you go

Pay less when you reserve Pay less when AWS grows

No up-front investment

We completed the equivalent

of thirty-nine years of

computational chemistry in just

under 9 hours for a cost of

around $4200.

Steve Litster

Global Head of Scientific Computing, Novartis

”

“

Novartis: Acceleration of pre-clinical R&D

• Existing infrastructure to screen 10

million compounds in a computational

model not available

• New infrastructure would have cost

approximately $40 million to build

Novartis used AWS for HPC

computational chemistry

Amazon EC2 purchasing options

On-Demand Reserved Spot

EC2 On-Demand pricing

Short-term, spiky and

unpredictable

Low cost and

flexibleDevelop and test

EC2 Reserved pricing

Upfront payments to

reduce costs

Steady state Optional capacity

reservationNew

Characteristic Standard

Payment No upfront

Partial upfront

All upfront

Commitment 1 year

3 year

Sellable on RI Marketplace Yes

Change Availability Zone, instance size

(Linux), networking type

YesConsole and API:

ModifyReservedInstances

Change instance families, operating

system, and tenancy

No

Savings* Up to 75%

Standard Reserved Instance details

* Dependent on AWS service, size/type, and region

2007 2008 2009 2010 2011 2012-2013 2014

m1.xlarge

m1.large

m1.small

m2.2xlarge

m2.4xlarge

c1.medium

c1.xlarge

m1.xlarge

m1.large

m1.small

cc2.8xlarge

cc1.4xlarge

cg1.4xlarge

t1.micro

m2.xlarge

m2.2xlarge

m2.4xlarge

c1.medium

c1.xlarge

m1.xlarge

m1.large

m1.small

c3.large

c3.xlarge

c3.2xlarge

c3.4xlarge

c3.8xlarge

cr1.8xlarge

hs1.8xlarge

m3.xlarge

m3.2xlarge

hi1.4xlarge

m1.medium

cc2.8xlarge

cc1.4xlarge

cg1.4xlarge

t1.micro

m2.xlarge

m2.2xlarge

m2.4xlarge

c1.medium

c1.xlarge

m1.xlarge

m1.large

m1.small

cc1.4xlarge

cg1.4xlarge

t1.micro

m2.xlarge

m2.2xlarge

m2.4xlarge

c1.medium

c1.xlarge

m1.xlarge

m1.large

m1.small

c3.large

c3.xlarge

c3.2xlarge

c3.4xlarge

c3.8xlarge

hs1.8xlarge

m3.xlarge

m3.2xlarge

hi1.4xlarge

m1.medium

cc2.8xlarge

cc1.4xlarge

cg1.4xlarge

t1.micro

m2.xlarge

m2.2xlarge

m2.4xlarge

c1.medium

c1.xlarge

m1.xlarge

m1.large

m1.small

c1.medium

c1.xlarge

m1.xlarge

m1.large

m1.small

new

existing

g2.2xlarge

m3.medium

m3.large

i2.large

i2.xlarge

i2.4xlarge

i2.8xlarge

r3.large

r3.xlarge

r3.2xlarge

r3.4xlarge

r3.8xlarge

t2.micro

t2.small

t2.medium

c4.large

c4.xlarge

c4.2xlarge

c4.4xlarge

c4.8xlarge

Increasing customer choice…

2010

latest

2015

c3.large

c3.xlarge

c3.2xlarge

c3.4xlarge

c3.8xlarge

hs1.8xlarge

m3.xlarge

m3.2xlarge

hi1.4xlarge

m1.medium

cc2.8xlarge

cc1.4xlarge

cg1.4xlarge

t1.micro

m2.xlarge

m2.2xlarge

m2.4xlarge

c1.medium

c1.xlarge

m1.xlarge

m1.large

m1.small

g2.2xlarge

m3.medium

m3.large

i2.large

i2.xlarge

i2.4xlarge

i2.8xlarge

r3.large

r3.xlarge

r3.2xlarge

r3.4xlarge

r3.8xlarge

t2.micro

t2.small

t2.medium

c4.large

c4.xlarge

c4.2xlarge

c4.4xlarge

c4.8xlarge

d2.xlarge

d2.2xlarge

d2.4xlarge

d2.8xlarge

m4.large

m4.xlarge

m4.2xlarge

m4.4xlarge

m4.10xlarge

t2.large

As Technology Advances…

…Convert Your Reserved Instances

The Convertible Reserved Instance is a new type of Reserved Instance

that can be exchanged during the 3 year term for new Convertible

Reserved Instances of equal or greater value. The new Convertible

Reserved Instances can correspond to a different instance family or a

new price, instance size, platform, or tenancy

instance optimized

instance

Convertible Reserved Instance details

* Dependent on AWS service, size/type, and region

Characteristic Standard Convertible

Payment No upfront

Partial upfront

All upfront

No upfront

Partial upfront

All upfront

Commitment 1 year

3 year 3 year

Sellable on RI Marketplace Yes Coming soon

Change Availability Zone, instance size

(Linux), networking type

YesConsole and API:

ModifyReservedInstances

Yes

Console and API:ExchangeReservedInstances

Change instance families, operating

system, and tenancy

No Yes

Savings* Up to 75% Up to 45%

EC2 Spot pricing

Users with urgent

computing needs or

large amounts of

additional capacity

Time or instance

flexible

Experiment and/or

build cost-sensitive

businesses

Spot Instance details

Options

• Spot Fleet to maintain instance

availability

• Spot block durations (1-6 hours) for

workloads that must run continuously

Commitment level

• None

* Compared to On-Demand price based on specific EC2 instance type, region, and Availability Zone

Spot rules

Markets where the price of compute changes based on supply and demand

You’ll never pay more than your bid

50% bid

of OD

75% bid of OD

25% bid of OD

You pay the

market price

87% discount!

Which EC2 purchasing model is right for me?

Have a balanced meal!

Why choose just one business model?

Use a combination of all three!

1. Use Reserved Instances for known/steady-state workloads

2. Set up multiple Auto Scaling groups

3. Scale using Spot, On-Demand, or both

0

2

4

6

8

10

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

/Spot

On-Demand

Spot

Reserved

Examples

Example: Three-tier web app

Tagging is essential!

• How do we explain the costs?

• How do we allocate the charges to the right team?

• How do we save money?

Tagging is essential!

Stateless

Scale based on demand

Web tier

0

5

10

15

20

25

30

35

40

0 2 4 6 8 10 12 2 4 6 8 10

Web servers

Reserved Spot On-Demand

0

50

100

150

200

250

300

350

400

0 2 4 6 8 10 12 2 4 6 8 10

Web traffic

App tier

Stateful

Scale based on demand

0

10

20

30

40

50

0 2 4 6 8 10 12 2 4 6 8 10

App servers

Reserved On-Demand Spot block0

100

200

300

400

500

0 2 4 6 8 10 12 2 4 6 8 10

App requests

Database Tier

Highly stateful

Static at peak load

0

0.5

1

1.5

2

2.5

3

3.5

0 2 4 6 8 10 12 2 4 6 8 10

Database servers

Reserved On-Demand Spot0

100

200

300

400

500

600

700

800

900

0 2 4 6 8 10 12 2 4 6 8 10

Database requests

Summary: Three-tier web app

0

10

20

30

40

50

60

70

80

90

0 2 4 6 8 10 12 2 4 6 8 10

Three-tier application servers

Reserved On-Demand Spot

Summary

Have a balanced meal! Across the three

tiers, our meal consists of:

• Spot 13%

• On-Demand 11%

• Reserved 76%

Remember!

“No server is easier to manage than no

server” - Werner Vogels, CTO, Amazon.com

Ubisoft uses AWS to develop and launch

social games quickly

• Ubisoft is a Paris-based gaming company, and

creator of popular gaming titles, including Assassins

Creed, Far Cry, and Just Dance

• Moving games to social and mobile platforms

required capacity to scale fast; using a traditional

environment would be an extensive and costly

investment

• Using the AWS Cloud to optimize games at the

application, caching, and data layers, improving the

user experience

”

“By using the AWS cloud we

were able to launch 10 social

games within 18 months.

Lenin Gali

Senior Director, Ubisoft

Example: Grid processing

Time

Typical server utilization rates are low due to need to deploy for peak needs…

The old way: Low utilization, high costs

Time

Higher grid utilization rates result in hidden costs: longer queue wait times

and delayed results

The old way: Managing utilization with grids

0

20

40

60

80

100

120

140

160

3 year Reserved Instances

?

The old way: In the cloud!

Higher utilization can reduce IT spending…

0%

20%

40%

60%

80%

100%

Grid Utilization

The world as seen by central IT

-600%

-500%

-400%

-300%

-200%

-100%

0%

100%

0

200

400

600

800

1000

1200

1400

Jobs in the Queue

Grid Utilization

But higher utilization also creates IT constraints…

Every job in the

queue represents

business impact

But Grid

Utilization looks

great!

The world as seen by the business

Reduced Time

Project

Acceleration

Scale higher to reduce time to results: shorter wait times, greater agility,

faster innovation cycles

The cloud way: Scalability when needed

?

The new way: In the cloud!

0

100

200

300

400

500

600

1 2 3 4 5 6 7 8 9 10 11 12

Optimizing for cost and business results

Spot

Spot block

On-Demand

3 year Reserved Instances

0

100

200

300

400

500

600

1 2 3 4 5 6 7 8 9 10 11 12

Going a step further with Spot blocks!

Spot

Spot block

On-Demand

3 year Reserved Instances

The new way: In the cloud!

?

“We constantly understate what our capabilities are to solve problems. The

biggest constraint is never the constraint of time or money, it’s generally the

constraint of thought.”

– Jeff Smith, CEO, Suncorp Business Services

Founded: 1996 • Employees: 15,000+ • Headquarters: Brisbane, Australia

Accelerating transformation

EC2 purchasing options by industry

Web scale (e.g. Adtech) company Enterprise SaaS company

Purchasing model by industry

Onboarding enterprise Gaming company

Purchasing model by industry

Technology companyScientific research

Purchasing model by industry

Different purchasing options in a single company

Data science

New app development Test and development

Internal IT

Let’s recap

Remember the pillars of optimization Right-sizing

Increase elasticity (turn stuff off!)

Measure, monitor, and improve

Use tags to understand your services

There are 3 core purchasing options – have a balanced meal

Architect your workloads with performance and cost in mind

Summary

Freedom to build

unfettered

Freedom to get real

value from dataFreedom to say yes

AWS is more cost-effective in both short-term and long-term than on-premises

environments. By leveraging the EC2 purchasing models, you gain the…

Thank you!

Remember to complete

your evaluations!

Related Sessions

CMP205: Lessons Learned from a Year of Using Spot Fleet

CMP302: Disrupting Big Data with Cost-effective Compute

CMP307: Save 90% and Run Production Workloads on SPOT

CMP316: Spark on AWS (with SPOT)

ENT209: Cost Optimization at Scale

WIN307: Searching Inside Video at Petabyte Scale Using Spot