1. Leveraging the Cloud for Green IT: Predicting the Energy, Cost & Performance of Cloud Computing © 2009 Optimal Innovations, Hyperformix & RS Performance Amy Spellmann,Optimal Innovations Richard Gimarc,Hyperformix, Inc. Mark Preston,RS Performance December 9, 2009 Session# 431

2. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 1 Leveraging the Cloud for Green IT: What are we going to talk about? We will describe & demonstrate a methodology for predicting performance, energy and cost for expanding on-premise IT into the Cloud. Cut through the hype:  View Cloud as a new platform in your infrastructure  Quantify the effect of utilizing Cloud Energy Cost Performance  Utilize a repeatable, scientific methodology

3. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 2 Leveraging the Cloud for Green IT: Agenda  Business Motivation  Cloud Considerations  Methodology  Case Study  Summary & Recommendations

4. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 3 Cloud Computing: What is it?  A general term for  Delivering hosted services over the Internet  Highly automated platforms that can be assigned to applications or services on demand  Services are broadly divided into 3 categories – Infrastructure-as-a-Service (IaaS) – Platform-as-a-Service (PaaS) – Software-as-a-Service (SaaS)  Cloud services have 3 distinct characteristics that differentiate it from traditional hosting – Sold on demand, e.g., by the minute or the hour – Elastic, a user can have as much or as little of a service as they want at any given time – Fully managed by the provider

5. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 4 Cloud Computing Promises  Flexible, on-demand IT services  Low entry cost (CapEx vs OpEx)  Timely  Scalable  Greener  Elastic Follow the herd blindly into the Cloud?

6. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 5 Is Cloud a Good Fit for You?  Have Cloud solutions evolved enough to meet the demands of your growing enterprise?  Is Cloud a good fit for your business from a cost, energy and performance perspective?  Are Cloud solutions greener? The decision requires a formal evaluation to ensure that the benefits can be realized

7. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 6 Energy Consumption: Are Clouds Greener? Energy consumption in the data center is often not well understood [Koomey] Economies of scale enable Cloud providers to be more efficient – Scalability and efficiency must be balanced to meet profit margins – Energy usage can be reduced through free cooling and specialty hardware (e.g. Amazon, IBM, Google, Microsoft) Cloud vendors must be able to demonstrate their energy efficiency

8. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 7 Why Consider Cloud Computing?  IT infrastructure at capacity, need to accommodate growth  Available power to the data center at capacity  Opportunity to expand on-premise IT with on-demand resourcing  Pay-as-you go cash outlay vs. large investment up front  Government & utility energy saving incentives: rebates, carbon penalties

9. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 8 Cloud Considerations AttributeTraditional On-PremiseCloud Computing Infrastructure Scalability  Server upgrade (CapEx)  Potential data center expansion (CapEx)  Cloud resources (OpEx)  Limits data center expansion (CapEx) Deployment Timeline  Weeks to months  On-demand in minutes/hours Infrastructure Management  Physical and virtualized components  (Virtual) Cloud resources Business Continuity  Responsibility of on-premise IT staff  Partner with Cloud provider Physical Resource Utilization  Size to handle peak loads  Low utilization during non-peak time  Scale dynamically with on-demand resources  Eliminates the need to overbuild (1 of 2)

10. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 9 Cloud Considerations AttributeTraditional On-PremiseCloud Computing Network Infrastructure  Data center IT staff responsible for networks and network expansion (CapEx)  Utilize Cloud network infrastructure (OpEx)  Additional on-premise network may be required (CapEx) Performance  Response time depends on workload volumes & supporting infrastructure  Dependent on responsiveness of the virtual resources  Potential new network delay between on-premise and Cloud Energy  Driven by IT infrastructure & supporting facility  Fixed limit on power draw  Energy consumption managed by the provider  Limits on-premise energy growth IT Budget Categories  CapEx for infrastructure  OpEx for facility & energy  Primarily OpEx for infrastructure (2 of 2)

11. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 10 Cloud Consideration – Set of Attributes Addressed in Case Study Infrastructure Scalability Physical Resource Utilization Network Infrastructure Performance Energy IT Budget Categories Not Addressed in Case Study  Deployment Timeline  Infrastructure Management  Business Continuity  Security  Software Licensing  Operations Staffing  Facilities Cost  Maintenance

12. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 11 Modeling & Capacity Planning Business Growth / Demand Decision Support for Leveraging the Cloud Predictive Performance & Capacity Management On-Premise Provisioning ? Cloud Provisioning ?

13. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 12 Methodology Traditional + Cloud Analysis Traditional Capacity Planning + Cloud Analysis Infrastructure Capacity Analysis Workload & Growth Analysis IT Provisioning Plan & Cost Optimization Business Planning & Communications Energy Footprint Impact Cloud Provider Capacity On-Premise + Cloud Performance Compare On-Premise vs. Cloud Identify Cloud Services

14. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 13 Case Study Data Center & Application Growth Challenges  Web-based eCommerce application  Current server infrastructure at capacity limit  Expecting 2x workload growth over next year  Data Center reaching power limits Capacity Planning Question: Should we expand our on-premise infrastructure to handle the expected growth, or should we leverage the Cloud to support the increased workload volume? On-Premise Provisioning ? Cloud Provisioning ?

15. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 14 Case Study Approach: Compare On-Premise vs. Cloud  Determine workload growth pattern  Size on-premise infrastructure to handle growth  Select a Cloud provider  Quantify required Cloud resources  Compare & contrast energy, cost & performance Capacity Planning Question: Should we expand our on-premise infrastructure to handle the expected growth, or should we leverage the Cloud to support the increased workload volume? On-Premise Provisioning ? Cloud Provisioning ?

16. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 15 Case Study: Capacity Planning Factors to Address  What additional on-premise infrastructure would we need to support workload growth?  What Cloud resources would we utilize to support the additional workload volume?  How does the cost compare; on-premise versus Cloud?  How is our energy bill affected by growing into the Cloud?  Is there a performance penalty for augmenting our on-premise IT infrastructure with the Cloud?  Are we able to leverage the on-demand nature of Cloud resources? These factors will enable us to answer:  Should we expand our on-premise infrastructure to handle the expected growth, or should we leverage the Cloud to support the increased workload volume?

17. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 16 Case Study Workload Growth Pattern  Historical data shows workload volume by hour  Simplify analysis by assuming 25% steps  Analyze 4 load levels instead of 24  Baseline workload: hours 0-5 & 22-23  Increased workload volume from hours 6-21; add on-premise capacity or route to the Cloud?

18. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 17 Case Study Today’s Current Baseline  27 servers  Maintain 70% utilization threshold

19. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 18 Case Study Size On-Premise to Handle 2x Growth  Increase workload by 2x  25% growth steps  Upgrade DB server  Add servers to other tiers (+31)

20. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 19 Case Study Select a Cloud Provider  Amazon Web Services (AWS)  Elastic Compute Cloud (EC2) - PaaS Standard Instances  High-CPU Instances  EC2 Compute Unit has the equivalent CPU capacity of a 1.0-1.2 GHz 2007 Opteron or 2007 Xeon processor

21. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 20 Case Study Prepare to Model Cloud Resources  Develop model representations of EC2 Standard Instances

22. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 21  Number of instances required to satisfy on-demand  3 instance sizes  Maintain 70% utilization threshold  Leave DB on-premise On-Demand Provisioning Case Study Quantify Required Cloud Resources

23. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 22  On-premise servers will handle workload from hours 0-5 and 22-23  Additional workload from hours 6-21 will be routed to the Cloud Case Study Cloud Workload Routing

24. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 23  Instance pricing per hour of use  Data transfer cost based on GB quantity of in/out per month Case Study Compute Cloud Cost per Month

25. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 24 Case Study Estimate Hourly Energy Footprint  Cloud EFP constant across the day – additional work routed to the cloud  On-Premise scenario EFP follows workload volume  +31 additional servers contribute to On-Premise EFP

26. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 25 Comparisons On-Premise vs. Cloud Scenarios  Energy  Cost  Performance Cloud provides a way to cap our on-premise energy consumption Cloud requires a lower initial investment, but costs accumulate over time Cloud may increase our business service response times

27. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 26 Cumulative monthly energy cost for the Cloud solution is approximately half of the On-Premise solution (29 vs. 59 servers) $1,602 vs. $3,092 per month Hour-by-hour energy consumption The peak on-premise EFP is 50% lower using the Cloud Comparisons Energy

28. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 27  Cloud scenario: on- premise servers handle all off-peak workload  On-premise upgraded servers built out to handle peak  Small Standard are slower than on-premise counterpart  Large & XLarge are faster than corresponding on- premise servers Comparisons Average Response Time: CPU  Workload routed to Cloud for hours 6-21  Cloud response time is weighted average of requests handled on-premise and those routed to the cloud

29. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 28  Keeping DB on-premise incurs additional network overhead  On average 1.5 DB accesses per transaction (max of 5)  Hybrid architectures may increase response time beyond acceptable limits  Chatty applications are not good candidates  Assume 100 msec round trip latency  Max response time increase > 0.5 seconds Comparisons Average Response Time: CPU + Network Latency

30. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 29 On-Premise  Purchase of 31 new servers  Energy for built-out infrastructure (58 servers) Cloud  Energy cost of on- premise servers (27)  On-demand monthly cost of Cloud instances and data transfer  Slope of Cloud curve is steeper, implying higher incremental cost (OpEx)  On-Premise OpEx is $3,098 - Cloud OpEx is $7,018 Breakeven Comparisons Cost of On-Premise vs. Cloud

31. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 30 On-Premise  Upfront CapEx investment for 31 new servers  OpEx at $3,098 per month for energy Cloud  On-demand Cloud requires no initial CapEx investment  Monthly OpEx of $7,018 Some Cloud providers now support both cash outlay timing models in which an upfront CapEx expense offsets lower recurring OpEx expenses. Comparisons Cost Breakdown $83,685$84,215

32. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 31 Summary Did we meet our data center energy challenges?  Cloud enabled us to keep our on- premise energy requirements flat  Dynamic provisioning enabled us to more closely match infrastructure & demand for energy

33. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 32  Hybrid models can suffer an increase in response time due to the Cloud instance interactions with the on-premise servers  There are two factors that affect response time: speed of the Cloud instances and additional network latency  Network latency can have a significant affect on client response time Summary Did we meet our performance requirements?

34. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 33 Summary Did we meet our data center budget requirements?  Growing into the Cloud requires less up-front (CapEx) investment  Cloud recurring cost can end up exceeding traditional data center cost

35. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 34 Summary Capacity Planning Question Should we expand our on-premise infrastructure to handle the expected growth, or should we leverage the Cloud to support the increased workload volume? Grow into the Cloud Utilize the on-demand Cloud to handle the expected 2x workload growth On-Premise energy usage will be effectively capped Defers need for data center expansion Eliminates CapEx for required On-Premise infrastructure  But  Cloud monthly expenses (OpEx) are larger than On-Premise: $7,018 vs. $3,098  Cumulative cost of Cloud may overtake on-premise solution  Cloud causes increase in average transaction response time  Capacity plan developed for 12-month a planning horizon

36. Leveraging the Cloud for Green IT: Predicting the Energy, Cost and Performance of Cloud Computing 35 Recommendations A Formal Methodology is Required  Focus your analysis on the Cloud provider’s pricing model(s)  Evaluate data transfer cost & any other miscellaneous costs carefully  Don’t ignore performance  Cloud response times must meet business requirements  Manage your energy footprint by leveraging the Cloud  Utilizing Cloud resources effectively caps your on-premise energy cost & consumption  Data center power efficiency can be managed with the Cloud  Balance CapEx & OpEx based on your business requirements  Cloud allows you to shift large upfront infrastructure investments into predictable recurring monthly costs

37. Blind Adoption Can Be Expensive Mitigate Your Risk With Proven Methodology & Best Practices Leveraging the Cloud for Green IT: Predicting the Energy, Cost & Performance of Cloud Computing