95% of your digital marketing spend is wasted. Here's how to fix it.
How to Find and Fix Waste to Optimize Your Cloud Spend
-
Upload
rightscale -
Category
Technology
-
view
153 -
download
5
Transcript of How to Find and Fix Waste to Optimize Your Cloud Spend
HOW TO FIND AND FIX WASTE TO
OPTIMIZE YOUR CLOUD SPEND
Panelists
• Kim Weins
• VP Marketing, RightScale
• Tim Miller
• VP Engineering, RightScale
• How Much Am I Wasting?
• Five Areas for Optimization
• Unused and Underused Resources
• Dev Environments
• Expensive Regions
• Old Instance Types
• Reserved Instances
• Prevention
Agenda
2
RightScale to Manage Any Resource Pool
Self-Service Cloud Analytics
Universal Cloud Management Platform
Cloud Management
Multi-Cloud Orchestration
3
Governance
Public
Clouds
Private
Clouds
Virtual
Servers
Bare Metal
Servers
18% 21%
24% 26%
2013 2014 2015 2016
Respondents that Say Managing Cloud Costs is a Significant Challenge
Challenge of Managing Costs is Growing
Source: RightScale 2016 State of the Cloud Report
3%
18%
17%
18%
21%
36%
29%
28%
28%
45%
4%
14%
22%
24%
25%
28%
31%
33%
34%
45%
Use Google Preemptible VMs
Use AWS Spot Instances
Move workloads to cheaper cloud/region
Select cloud or region based on cost
Track AWS Ris to make sure they are used
Purchasing AWS Ris
Look for storage volumes not in active use
Shutdown workloads during certain hours
Automate shutdown of temporary workloads
Monitor utilization and rightsize instances
How Companies are Optimizing Cloud Costs
Enterprise
SMB
Few Companies are Optimizing Cloud Costs
Source: RightScale 2016 State of the Cloud Report
DOWN
OFF
Monthly Spend Savings Identified
Customer #1 $200,000 44%
Customer #2 $26,000 40%
Customer #3 $46,000 33%
Customer #4 $19,000 34%
Customer #5 $137,000 20%
Our Last 5 Cost Optimization Reviews
Cost Review Example
RightScale Example
UNUSED AND
UNDERUSED
The High Cost of Overprovisioning
m3.xlarge
$.266
$2330/year m3.large
$.133
$1165/year m3.medium
$.067
$587/year
Save
50% Save
75%
Underutilization is Rampant
Memory utilization
CP
U u
tiliz
ation
20-40%: Size -1
<20%: Size -2
0
10
20
30
40
50
60
70
80
90
100CPU Util%
Mem Util%
Don’t Provision to the Peak
Don’t provision to the peak!
Provision for normal loads
and auto-scale for peak
Don’t Forget Storage
Volumes
Snapshots
Attached Unattached
Finding Unattached Volumes
Cost • AWS = $200+ per month
• Azure = $40-120 per month
• Google = $80-340 per month
DEV ENVIRONMENTS
Development Environment Usage Hours
Mon Tue Wed Thu Fri Sat Sun
00:00
11:59
24x7
168 hours
12x5
60 hours
35%
Development Environment Shutdown Dates
Needed
for 3 days
Left running for 4 days
Left running for 7 days
Left running for 14 days
25% waste
57% waste
79% waste
A Little Waste Adds Up
Typical m3.large ($.133)
24x7
16 days
Optimized
m3.medium ($.067)
12x5
14 days
$51.07
$11.26
78%
less
$102,140
$22,520
Per launch 100 devs * 20x/yr
Save
$79K
Scheduling Workloads in RightScale
• Training
• Demos
• Sandboxes
• Test
• Staging
Other Temporary Workloads
EXPENSIVE REGIONS
Regional Differences in AWS Example
Region Location
Instance
Size
Hourly
Cost
Cheaper
Region Location
Hourly
Cost % savings
us-west-1 NorCal m3.large $0.15 us-west-2 Oregon $0.13 14%
eu-central-1 Frankfurt m3.large $0.16 eu-west-1 Ireland $0.15 8%
ap-southeast-1 Singapore m3.large $0.20 ap-southeast-2 Sydney $0.19 5%
ap-northeast-1 Tokyo m4.large $0.17 ap-northeast-2 Seoul $0.17 5%
Expensive
Region
Monthly
Spend
Cheaper
Region
%
savings
Monthly
savings
us-west-1 $5,000 us-west-2 14% $700
eu-central-1 $0 eu-west-1 8% $0
ap-southeast-1 $2,000 ap-southeast-2 5% $100
ap-northeast-1 $0 ap-northeast-2 5% $0
$800
Regional Differences in Azure Example
Region Location
Instance
Size
(Linux)
Hourly
Cost Cheaper Region Location
Hourly
Cost
%
savings
East US Virginia D1v2 $0.07 East US 2 Virginia $0.06 12%
North Central US Illinois D1v2 $0.07 South/West Central US Texas $0.06 12%
Central US Iowa D1v2 $0.07 South/West Central US Texas $0.06 12%
West US California D1v2 $0.07 West US 2 $0.06 12%
Canada Central Toronto D1v2 $0.08 Canada East Quebec City $0.07 9%
West Europe Netherlands D1v2 $0.08 North Europe Ireland $0.07 14%
East Asia Hong Kong D1v2 $0.11 Southeast Asia Singapore $0.09 15%
Japan East Tokyo D1v2 $0.11 Japan West Osaka $0.09 13%
Australia East NSW D1v2 $0.09 Australia Southeast Victoria $0.08 7%
OLD INSTANCE TYPES
Old Instance Types
Cloud Provider
Previous
Generation
Current
Generation Size mapping
% savings
(East)
AWS T1 T2 same size 35%
AWS M1 M3 same size 24%
AWS M2 R3 downsize one 32%
AWS C1 C3 same size 60%
AWS CR1 R3 same size 24%
AWS HS1 D2 downsize one 40%
Azure D* D*v2 same size 4%
RESERVED INSTANCES
• We know I can save with RIs, BUT:
• I don’t have time to analyze it
• I know I have underutilized instances, so I don’t want to buy RIs on
them
• I’m implementing Docker and that will change what I need
• We have Dev instances that are changing all the time
• I need to re-architect that system
• I may need to change my instance sizes
• There may be new instance types coming
• ….and more
How to Cope with the “Buts” of RIs
Five Principles of Managing RIs
Think “RI coverage” % of instances that are covered by RIs
Aggregate and share RIs Link accounts in AWS
Select right coverage level More variability = lower % coverage
Plan to Modify RIs Track utilization and change RIs as you go
Sell RIs If you can’t modify, then sell
Sharing RIs Across Accounts in AWS
AWS Payer
Account
Linked
Account
Linked
Account
Linked
Account
Independent AWS
Account
RI
Unblended costs
1. RI will be applied in account where purchased first
2. If no matching instances, it will be allocated to other
accounts in the family
Blended costs
• The savings from the RI gets shared proportionally
based on usage of instances that match the RI.
RIs will be shared = more flexibility = higher coverage RIs NOT shared
Set up
consolidated
billing
Reserved Instance Coverage
100 instances
50 Reserved Instances
50% RI coverage 50% On-Demand pricing
Example: Target RI Coverage by Usage Model
Target RI Coverage: 75-85%
Production
Instances
Target RI Coverage: 75-85%
Dev
Instances
Example: RI Coverage for Low Utilization
100 medium instances
100 large instances
Target RI Coverage 40%
Today 50% of instances
have low utilization
Later We’ve downsized
instances, modified
RIs
Modify RIs (1 large = 2 medium), coverage now 80%
RI Factors Modification? Notes
Instance family/size Size only, not family Only for Linux (not
Windows, RHEL,
SUSE)
Region/AZ AZ in same region Depends on capacity
OS No
Network (Classic/VPC) Yes
Rules for Modifying Reserved Instances
Must Maintain “Footprint” When Modifying RIs
xlarge
large
large
medium medium
medium medium
large
medium medium
Original
footprint Option 1 Option 2 Option 3
Example: Reserved Instance Utilization
Automate to Prevent Drift
37
Now RI Utilization Analysis and
Modifications
Ongoing Set up alerts for
underutilized RIs
Now Find unused instances
Ongoing Schedules, automated
alerts, terminator scripts
Now Check budget vs. actual
Ongoing Set up budget alerts