Key Considerations in Planning Performance & Capacity in Virtualized Environments

This presentation is among the Top 27 Best Papers/Practice/Tutorials selected, out of 460+ submissions received, to be presented @STC 2012.

Presentation Abstract

Virtualization has become the buzz-word for Information Technology departments. IT departments have always had the need to do a lot more with lot less at their disposal. Companies who plan to move to server virtualization gain many benefits; most notably cost savings on hardware, power & cooling, as well as flexible response to peak demands. Server virtualization has lot of inherent benefits – cost savings, server consolidation etc. but one should be very tactful in managing Capacity & Performance of the virtualized server farm. It is important to note that this adds another layer to the already existing complex IT infrastructure. Needless to say, the gains of virtualization will be lost if applications deployed on the Virtual servers become sluggish & end-users complain of long-running response times from an end-user perspective.

About the Author

V Satya Prakash has an overall 10+ years IT experience out of which 9+ years of extensive experience on Performance Testing, Performance Benchmarking for OLAP/OLTP systems Proficient with Load Testing tools - Load Runner, QA Load, Silk Performer, Web Load & Open STA Expertise on Performance Tuning - strengths on Oracle SQL tuning, Stats Pack utility, Explain Plan & DB2 Performance analysis Bachelor’s Degree in Engineering
Worked on Virtualization Load testing strategies for various pursuits.

Transcript

1. Key considerations in Planning Performance & Capacity in Virtualized ©2012 Deloitte. All rights reserved 1 Capacity in Virtualized Environments - Satya Prakash
2. Abstract –Server virtualization has lot of inherent benefits – cost savings, server consolidation etc. but one should be tactful when it comes to managing a complex datacenter – when it comes to addressing 2 key aspects – Capacity – Performance –IT departments have always had the need to do a lot more with lot less at their disposal. Companies who plan to move to server virtualization gain many ©2012 Deloitte. All rights reserved 2 their disposal. Companies who plan to move to server virtualization gain many benefits; most notably cost savings on hardware, power & cooling, as well as flexible response to peak demands –But it is important to note that this adds another layer to the already existing complex IT infrastructure. Needless to say, the gains of virtualization will be lost if applications deployed on the Virtual servers become sluggish & end-users complain of long-running response times from an end-user perspective
3. 1.Challenges 2.Solutions 3.Key takeaways ©2012 Deloitte. All rights reserved 3 3.Key takeaways
4. Platform 1 machine, 1 OS, 2+ applications running Any change in the platform, either the app or OS would lead to a totally different environment set-up App-OS No. of applications deployed under an OS “Virtualization itself is the 1st load test!” Few Drivers Nature of challenges Challenges ©2012 Deloitte. All rights reserved 4 App-OS deployment “Virtualization itself is the 1 load test!” Host/ Guest? What is your guest? What is your host? Host= Windows; Guest= Linux & vice-versa is different Resources that guest OS is allowed to use matters Resource Usage Guest OS may report high CPU/ Memory usage But Host may still have idle capacity Only server monitoring is not a viable strategy
5. Challenges (Few illustrations) ©2012 Deloitte. All rights reserved 5 Physical Server layout Virtualization layout Complicated I/O problem!
6. Challenges (Few illustrations) ©2012 Deloitte. All rights reserved 6 Important to isolate the problem layer in the “Virtualization Stack”
7. Solutions/ Best ©2012 Deloitte. All rights reserved 7 Solutions/ Best practices
8. These methods simply don’t see the end-to-end Monitoring CPU, Memory: They may give some The only effective way to troubleshoot application performance issues in a virtualized environment is with an end-to-end analysis that starts from the end-user perspective Bottom-up Appproach: Component-based monitoring tools, software instrumentation, log files and tools from virtualization vendors Conventional Monitoring Consequence Method 1. Analyze System in an efficient way ©2012 Deloitte. All rights reserved 8 These methods simply don’t see the end-to-end application flow with virtual servers being dynamically provisioned and de-provisioned Monitoring CPU, Memory: They may give some visibility into the performance of the virtual layer, but none inside the applications running in the layer. Some of the important metrics as we would’ve seen in the load testing world — such as CPU usage, network load and disk I/O — are meaningless in a virtualized world Server Monitoring Remember: Capacity is not a separate silo from Performance!
9. 2. Do not neglect Trends. Consequently forecast Capacity ©2012 Deloitte. All rights reserved 9 Fig: Utilization trends
10. 3. Accommodate Trend and forecast Most important questions to answer: Add another host Add CPU to ALL hosts Add Memory to ALL hosts Add additional data-stores & spreading the Virtual Machines across Even decreasing forecasts should be accounted! ©2012 Deloitte. All rights reserved 10
11. Some of the key questions you want to answer in the forecasting exercise are: • How many objects will I be consuming in the future? • Will I need more capacity in the future? • When will I fill my current capacity? On the topic of making “Adjustments” to Capacity, some of the important scenarios that arise are as below: • Adding another host to a cluster 3. Accommodate Trend and forecast (continued…) ©2012 Deloitte. All rights reserved 11 • Adding more memory to hosts in a cluster • Adding additional CPU to hosts in the cluster • Adding another data-store and spreading VM disks across the additional data-store • Increasing CPU, memory reservations or priority on a resource pool
12. 4. Scalability –Database applications running on Physical servers become prime candidates for server consolidation since DB workloads are often provisioned on systems with high CPU & Memory configurations. –Aggregate TPM scales in a nearly linear fashion as VM’s are added. The scaling tapers off only towards the end because at that stage the resources (especially CPU) are nearing saturation ©2012 Deloitte. All rights reserved 12
13. 4. Scalability (continued…) Higher TPM the better –There is not much difference in terms of transactions per minute for an ESX server vs. Physical server ©2012 Deloitte. All rights reserved 13 Average Transaction response times show a difference of ~13% for standard Query Performance benchmarks
14. Bottlenecks –Bottlenecks exist in any system – be it on a Physical Environment. With Virtual Machines, need all the more details since there’s an additional complexity with the infrastructure itself. Measure & monitor the host OS & at the application layer for each VM Need to differentiate between VM’s (VM to VM) bottlenecks vs. System bottlenecks ©2012 Deloitte. All rights reserved 14 –Few “throughput limitation” bottlenecks in general Disk I/O when Virtual machines used as Load Generators Shared Disk resources for multiple applications Virtual Memory Paging, especially on shared disks CPU time shared between Processor intensive applications Multiple Virtual machines using the same Network Interface Card to transfer data
15. Key takeaways ©2012 Deloitte. All rights reserved 15 Key takeaways
16. Capacity –Capacity is NOT about how much you can cram into your Virtual infrastructure –Capacity is more than just applying mathematical calculations; you need best practice intelligence and logic Performance Performance Bottlenecks ALWAYS exist. The key questions to understand & answer are: ©2012 Deloitte. All rights reserved 16 answer are: –When will I encounter this bottleneck? –What will this bottleneck impact? –When do I need to start really caring about this bottleneck? –What will I need to do to alleviate this bottleneck?
17. Contact Satya Prakash ©2012 Deloitte. All rights reserved 17