aws re:invent 2016: start your cost optimization program: learning from intuit’s cloud...
TRANSCRIPT
© 2016, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Elvy Lee, Sr. Finance Leader, Intuit
Dieter Matzion, Sr. Business Analyst, Intuit
Keith Jarrett, Business Development Manager, Cloud Economics
December 1, 2016
How to Start Your Own Cost
Optimization ProgramLearnings From Intuit’s Cloud Journey
ENT315
What We Learned.
• Savings potential is large!
• Framework for implementing best
practices at scale
• Architecture best practices
• Ongoing operational best practices
• At the center of cost optimization is a
team
What to Expect.
• Recapping Cost Optimization Pillars
• Implementing Best Practices
• Establishing Metrics
• Creating a Cloud Culture
• Migrating to the cloud
Limit
Provisioning
Target
Setting
Automation
Periodic
Review
Charge-
backs
Pricing
Model
Autoscaling
Turn off
unused
Cost
Optimized
Direct
Inputs
Indirect
Inputs
Key Cost
Optimization
Drivers at
Scale
The Five Pillars of Cost Optimization
Right-Sizing Your
Instances
Pick the Right
Pricing Model
Increase
Elasticity
Measuring &
Monitoring
Match Usage to
Storage Class
© 2016, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Elvy Lee, Sr. Finance Leader, Intuit
Dieter Matzion, Sr. Business Analyst, Intuit
December 1, 2016
How to Start Your Own Cost
Optimization ProgramLearnings From Intuit’s Cloud Journey
ENT315
What to Expect from the Session
Intuit overview and journey to AWS
Share how we got started on cost optimization … initial focus on reserved instances (RIs)
External best practices
Active management of RIs
Automation, reporting, and mechanisms to ensure low waste, high leverage
Go through the next set of tools and capabilities we’re building
Ops and tools relevant to us
Driving a cost-conscious culture through partnering and chargebacks
Questions, suggestions, and ideas along the way and at the end
Our goal: We learn something from each other
1
2
3
4
Why AWS & the public cloud?
• Enable speed & ease for our developers
• Accelerate towards being true cloud & SaaS
• Scalable, elastic, global and reliable
Why an optimization program?
• Lots of opportunity to tune costs
• Best-in-class, large scale SaaS companies
have focused teams on this
• We are still discovering all the levers we have
Made significant progress; much more to accomplish
Public Cloud
Intuit Hosting
• Multiple data centers • Consolidated & fewer data
centers
A FEW YEARS AGO … TODAY
• Monolithic code bases
• Manual deployment and
configuration
• Cloud-based services
• Continuous Integration /
Continuous Deployment
Uncovering New Opportunities for Hosting
AWS SPEND RAMPING UP
FY’17FY’16FY’15FY’14 FY’18
STRATEGY
1. Reserved instances (RI) – Prepay for
capacity to realize up to 60% savings
vs. on-demand (OD) prices
2. Elasticity – Turn off non-production
instances during off-hours; don’t
establish a compute footprint that’s
larger than customer demand dictates
(pay only what you use)
3. Right-sizing – Use most appropriate
instance family/type for the workload
Thoughtful but Scrappy Start to Our Cost Optimization
External learnings Learnings & pivotsOngoing metrics &
mechanisms
Quick experiment: Small RI
purchases to get operational
understanding
Learnings & pivots:
• Transparency, reporting, and tools
were an important initial investment
• Insufficient capacity errors & reporting
• Lots of non-standard instances …
need on-going hygiene & clean-up
plan
• Needed to leverage automation
• Evangelize the teams … 2-3 six-
pagers to align on strategy, benefits,
risks and best practices
Completed best-practices AWS check
with SaaS technology peers
• Ongoing mechanisms with CFO,
CTO, our “cloud stewards” to
ensure progress
• Monthly purchases & modification
• Success measures:
• RI coverage 90%+ “standard”
instances
• Waste <10%
• Cumulative savings measures
INITIAL OPPORTUNITIES
Leverage reserved instances (RI)
Transparency & tools … create the
data and reports so we buy the right
stuff
Establish mechanisms & purchase
cadence (every month)
Establish scrappy central team …
to drive execution across
2
3
4
1
External learnings Learnings & pivots
Impact of implementing reserved instances
• Central RI strategy
• Usage clean-up plans
• Monitor & analytic tools
• Fragmented setup
• Diverse instances
• No best practice sharing
FROM TO
Total usage
Good Progress, More Work to Do
Today
No active management of RIs
No automation, no reporting,
14-day purchase cycles
No deployment standards, no policies,
no best practices
Lagging industry on cost optimization
Last Year
Ongoing program, saving dollars to be re-invested
in engineering and product
RI coverage from ~20% to ~80% (goal = 90%)
Joint team with Finance & Product Infrastructure
Several dashboards; improved data ingestion of
AWS feeds (e.g., DBR, ICE, VPC mapping)
4-7 day purchase cycles (goal = 1 day)
Designed and implemented a “clean-up” plan
Catching up fast!
Limited AWS ops involvement Several white papers informing roadmap & strategy
Continue Enhancing Our AWS Ops
• Initial focus is visibility into AWS, starting with DBR, ICE and utilization, and
moving to code tracing and systems performance (e.g., performance counters)
• More complex systems like purchase recommendations, right-sizing, A/B cloud
impact, and squeeze testing, will be fed from the systems that provide visibility
• Establish deeper partnership with service owners to guide consistency
Squeeze Testing
Detailed Billing
Report (DBR)PerformanceUtilization Code Tracing
RI Purchase
Recommendation
RI Modification Right-sizing A/B Cloud Impact
Co
st
Op
tim
iza
tio
n
Operational & Technical Capability
We’ll Be Focusing in These Core Areas in the
Coming Months
• Shorter RI modification times …
automate modifications and purchases
• Complete ”clean-up” plans
• Track purchases, build RI database
• Self-service reporting for all AWS data
feeds
• Continue to actively track: (1) Amazon
EC2 launch events, (2) insufficient
capacity errors (ICE), (3) behavioral
changes of users
• Build near real-time ICE feed using
AWS CloudTrail
• Utilization (CPU, RAM, disk, network):
build a right-sizing framework
• Track Amazon S3, Amazon RDS,
Amazon EBS, Snapshots: automated
cleanup, lifecycle management
• Turn off dev/test environments outside
of business hours
• Scale back bigger offerings during
non-peak
Building a Cost-Conscious & Active
Management Culture
PARTNERING CHARGEBACKS TO DRIVE ACCOUNTABILITY
• Help teams with migration planning
• Work with cloud stewards to
understand “click-down” plans
• Share best practices, standards,
and any clean-up initiatives
We pass through savings & costs to business unit
• We charge the business P&Ls for every unit of
hosting (internal and external)
• Chargeback for AWS costs
• Pass back RI savings to teams using a blended
rate approach
Chargebacks must:
• Drive accountability and incent the right behaviors
• Be transparent and easy to understand and
implement
• Meet SEC reporting requirements
In Closing …
• Start small, but just get going!
• Invest in reporting transparency and visibility
• Metrics, mechanisms, and chargebacks to drive good
behaviors … passing back both costs and savings
• Reinforce an active management mindset