2. About Me
Cloud and Enterprise IT Entrepreneur
CEO, CloudVertical - “Reducing IT Waste”
Previously GM, Hosting365 (now SunGard)
@edbyrne
www.edbyrne.me
blog.cloudvertical.com
3. Agenda
What is Cloud Economics
The Myths of Cloud Computing
Variables and Cost Models
Resources
4. Principles of Cloud Economics
IT as a Utility - Available and Scalable
IT is Paid for using a Consumption based model
Theoretically 100% capital efficient
5. IT Cost Model
Capacity is CapEx and Cyclically Driven
Usage follows predictable patterns
Cost increases with time and complexity
6. Cloud Cost Model (Theory)
Capacity scales up and down in line with demand
Usage follows predictable patterns (no change)
Cost matches capacity - pay as you consume
7. IT Cloud Cost Model (Actual)
Capacity is provisioned based on expected load
Usage follows predictable patterns
Cost goes up and to the right, with an **** moment
8. Cloud Economics Reality
The Cloud is not analogous to the Electricity model.
The Cloud is not Pay as You Go. It’s Pay as You Provision.
Utilisation is the problem -> the traditional capacity planning
model has migrated to the Cloud.
• Physical Hardware - 20% Utilisation Average
• Virtualised Infrastructure - 30% Utilisation Average
• Cloud Deployments - 30% Utilisation Average
10. Traditional Culture
IT has never been responsible for Cost metering
Budgets are aspirational targets
Existing tools focus on Uptime, Usage and Performance Monitoring
11. Efficient Computing
You can’t manage what you don’t measure
Create Accountability, Visibility, Reporting
Meter Cost, Capacity and Usage
Show-back reports for each use type
1. Inform 2. Improve 3. Maintain
14. Truth
Like for Like - More Expensive
But it’s not Like-for-Like
• Scalability & Flexibility
• Resilience & Redundancy
• OpEx not CapEx
More Cost Effective for a lot of workloads
16. Truth
Versus Fully Loaded Server Cost - True 90% of time
Do you know the true costs of your IT infrastructure?
Cloud often ‘looks’ more expensive but is not.
Netflix vs Zynga models -> both right
24. Truth
Jevon's Paradox : The easier it becomes to consume a
resource, the more of it that will be consumed.
The Cloud has enormous efficiency capabilities.
Waste is not a technology problem, it’s a people one.
25. AWS Costs Cheat Sheet
Know the Units Instance Storage
EBS Standard
EBS PIOPS
EBS Requests
Storage
N/A
10c /gb-month (provisioned)
12.5c /gb-month (provisioned)
10c per million
Regional Cost Differences
Region
US East (Virginia)
US West (Oregon)
US West (California)
Variation
BASE
0%
12.5%
EBS Provisioned IOPS 10c /month (2.628m requests) EU (Ireland) 8%
Compute - CPU/hour EBS Snapshot
S3 Standard
S3 Reduced Redundancy
12.5c /gb-month (stored)
12.5c /gb-month (stored)
9.3c /gb-month (stored)
Asia (Singapore)
Asia (Toyko)
South America (Sao Paulo)
8%
15%
45%
Memory - GB/hour
Glacier 1c /gb-month (stored)
Instance Sizes Network
API name Compute Memory Hourly Cost Data IN FREE
Storage - GB/stored
t1.mirco <2 0.613 0.02 Data OUT 12c /gb
m1.small 1 1.7 0.08 Data within AZ FREE
m1.medium 2 3.75 0.16 Data within Region 1c /gb
m1.large 4 7.5 0.32 Managed Data (EIP/ELB) 1c /gb
Network - GB/managed
m1.xlarge 8 15 0.64 Load Balancer 2.5c /hour
m2.xlarge 6.5 17.1 0.45 Load Balanced Traffic .8c /gb
m2.2xlarge 13 34.2 0.9 IP Address Per Instance FREE
m2.4xlarge 26 68.4 1.8 Extra or Unused IP Address .5c /hour
c1.medium 5 1.7 0.165
c1.xlarge 20 7 0.66
Time Key
cc1.4xlarge 33.5 23 1.3 Day = 24 Hours
24 hours /day
cc2.8xlarge 88 60.5 2.4 Week = 168 Hours
cg1.4xlarge 33.5 22 2.1 Month = 730 Hours
hi1.4xlarge 35 60.5 3.1 Year = 8,760 Hours
Month = 2,628,000 Seconds
168 hours /week Micro
Instance
On-Demand Instance Costs
Hour
0.02
Day
0.48
Week
3.36
Month
15
Year
175
730 hours /month
Small 0.08 1.92 13.44 58 701
Medium 0.16 3.84 26.88 117 1402
Large 0.32 7.68 53.76 234 2803
Extra Large 0.64 15.36 107.52 467 5606
8,760 hours /year Extra Large High Memory
Double Extra Large High Memory
Quadruple Extra Large High Memory
Medium High CPU
0.45
0.9
1.8
0.165
10.8
21.6
43.2
3.96
75.6
151.2
302.4
27.72
329
657
1314
120
3942
7884
15768
1445
Extra Large High CPU 0.66 15.84 110.88 482 5782
Quadruple Extra Large Cluster Compute 1.3 31.2 218.4 949 11388
Eight Extra Large Cluster Compute 2.4 57.6 403.2 1752 21024
40 hours working week Quadruple Extra Large Cluster GPU
Quadruple Extra Large High I/O SSD
2.1
3.1
50.4
74.4
352.8
520.8
1533
2263
18396
27156
173 hours working month Light
Average Reserved Instance Savings
Type % Saving Yr 1
40
% Saving Yr 3
56 All Prices in Dollars, from US East Virginia as published
2,076 hours working year Medium
Heavy
48
53
65
70
11 October 2012. Tiered Discounts and Free Tier Amounts
Not Applied. For more info go to www.cloudvertical.com
26. Optimise - Rightsize
Development Environment : 40 hours per week (24% available time)
Load Test / System Patch / Software Upgrade: 8 hours / 1 day - once-off
Back-office Applications : 40 hours per week ‘heavy’; outside hours 25% capacity
Travel / E-Commerce Site : Busy Consumer Hours. Seasonal. Marketing Driven.
27. Resources www.cloudvertical.com
Deck: blog.cloudvertical.com
AWS Economics Centre : http://aws.amazon.com/economics/
Cloudonomics Book: http://www.amazon.com/Cloudonomics-Website-Business-Value-Computing/dp/1118229967
Microsoft Cloud Economics: http://www.microsoft.com/en-us/news/presskits/cloud/docs/the-economics-of-the-cloud.pdf
AWS Costs Cheat Sheet: https://blog.cloudvertical.com/2012/10/aws-cost-cheat-sheet-2/
On-Premise vs. Cloud Cost Model: http://andrewmcafee.org/2012/10/mcafee-cloud-costs-google-model
Data Gravity: http://datagravity.org/