Use the Capacity Planning worksheet to do your production capacity planning. This worksheet requires information from the initial requirements plan and your actual performance test results. For more information on capacity planning and examples using this worksheet, please refer to Chapter 15. Part 1: Requirements SummaryPart 1 of the Capacity Planning worksheet contains a summary of the key requirements for your web site: load, throughput, and response time. Use this data to help determine your hardware needs. Part 2: Performance Results Summary
Part 2 contains a summary of the final performance test results. Use the final, tuned results from the performance worksheets recorded during the test. Appendix D provides example performance worksheets. Single-Server ResultUsing the performance worksheet or graphs of your test results, locate the performance of a single server corresponding to your response time objective, as specified in line 2 above.
Cluster Scalability ResultsUsing the performance worksheet or graphs of your cluster test results, locate the performance of the cluster corresponding to your response time objective, as specified in line 2 above. We recommend doing at least a three-server cluster test. Add more rows to this table for larger tests.
CPU Utilization DataFrom the performance worksheet, locate the CPU utilization for each of the servers during your cluster runs. Make sure to record the CPU utilization for all servers in the test bed. Add more columns for additional servers and more rows for larger cluster tests.
Part 3: Capacity Planning EstimatesThis section estimates hardware requirements for your web site using the data from Parts 1 and 2. Cluster Scaling RatioCalculate the scaling achieved during your cluster runs. You do this by calculating the ratio between the throughput of the cluster and the individual server's performance.
Estimate Scaling RatioThis step estimates the scaling ratio as we add additional servers to the cluster. This section of the worksheet assumes that additional servers follow the same scaling pattern measured in the lab. We do not recommend estimating beyond twice the tested performance. For example, to calculate the scaling ratio for four application servers, take the scaling ratio for three application servers plus the delta between the ratio at three application servers and two application servers. This sum gives us the projected scaling for four servers. Generally, the calculation for the scaling ratio for n number of servers is ratio ( n “ 1) + (ratio ( n “ 1) “ ratio ( n “ 2 )).
Estimate Throughput and LoadThis step estimates the concurrent users and throughput as you add additional servers to a cluster. The estimated load is a pure linear projection. Derive the estimated throughput using the scaling ratio calculated above. Continue adding servers to your estimate until you meet both the load and throughput requirements.
Estimate CPU UtilizationTo complete the hardware estimates, consider CPU utilization on all the servers included in your web site. You cannot achieve the application server scaling estimates if other servers' CPU utilization becomes a bottleneck. CPU Utilization per Additional ServerCalculate the increase in CPU utilization for each type of server as you add more application servers to the test cluster.
HTTP Server CPU Utilization ProjectionsEstimate the number of HTTP servers required to support the projected load and throughput. This calculation takes the CPU utilization of the preceding cluster run and adds the delta CPU utilization increase anticipated as both the load increases and the cluster adds an additional application server. Once the CPU utilization projections reach 80% or higher, plan to add an additional HTTP server, or to add additional processing power to existing server(s).
Database Server CPU Utilization ProjectionsEstimate the number of database servers required to support the projected load and throughput. This calculation takes the CPU utilization of the preceding cluster runs and adds the delta CPU utilization increase anticipated as the load increases and the cluster adds an application server. Once the projected CPU utilization reaches 80% or higher, plan to split your databases across additional servers or add more processing power to the existing server.
Other Server Utilization ProjectionsFollow the same methodology used above for all other servers in your environment.
Summary CPU Utilization Data
|