Capacity Planning

Suggested Capacity Planning Methods

Designing a site for commerce is more complex and difficult than designing for the other scenarios presented here. Here are some guidelines:

1. Prior to purchasing any equipment, create a logical diagram of the site and identify all components and major systems with which the Web site must interact.
2. Determine how many transactions per second the Web site will be expected to handle during its peak traffic time.
3. Using the information from steps 1 and 2, determine which components will be the most frequently used and rate them in terms of their use of server resources. (It is probably sufficient, at this stage, to focus on three parameters: connections per second, CPU utilization, and memory utilization.)
4. As the components are installed, test each of them by running simulated traffic into the component and recording how many transactions per second it can handle. At the same time, observe (using PerfMon, which is a performance monitoring tool that measures memory use system-wide) how much of the system’s resources are used by each component.
5. Using the information from steps 3 and 4, create a table similar to Table 4.5 (the numbers in this example are arbitrary):

   Table 4.5   Transaction Component Performance Comparison

   Component Name	Component “Rating”	Peak Transactions Expected	Maximum Transactions per Server	Server Resources Used per Transaction
   Sales Tax	1	200	70	1.4%
   Inventory	2	150	120	0.23%
   Customer Service	3	80	100	0.45%

   
   

6. Using the table you generated in step 5, determine which components are the most resource-intensive, and whether any will fail to deliver the peak transaction numbers expected, on a per-server basis.
7. Multiply the maximum simultaneous transactions per server by the server resources utilized per transaction to find the sum result for all components that run on the same server. Any total that exceeds about 70 percent indicates that you should either content stripe by adding servers, separate components onto their own servers, or both.
8. As the Web site is deployed, analyze periodic PerfMon performance reports on metrics like connections per second, memory, and CPU utilization to determine how accurate your chart is.
9. Update the chart as necessary with production data and add additional servers (using either component striping or specialization) to maintain the Web site, so that the peak transactions per second will only use approximately 70 percent of the resources of any one server computer.

© 1997-1999 Microsoft Corporation. All rights reserved.