Benchmarking Server Performance, Part 2
by Heidi Brumbaugh
How Do We Measure Server Performance?
When we turn to analyzing Web server performance, deciding what to measure is certainly
the first step in benchmarking. Right now the leading benchmarks tend to focus on two
measurements of performance: throughput-the rate at which the server can process HTTP
requests; and response time-the time a server spends processing a single request. A typical
approach is to configure the benchmarking software to simulate a large number of clients,
and then to request a set of pages of varying lengths, so you can see results for both small
and large files.
To give you an idea of what kind of results you'd be looking at, just below are
SPECweb96's benchmark results for Digital's Alphaserver 4000 5/400 running Zeus
1.1.5. Full details of this performance evaluation are at open.specbench.org/osg/web/results/res96q4/web96-961125-
01539.html
(see below for a description of SPECweb96).
Throughput Response
(ops/sec) (msec)
117 9.0
234 8.9
351 8.9
468 9.4
586 10.5
703 11.5
820 13.1
937 15.4
1054 19.8
1157 30.2
There are a number of Web server benchmarking programs available. One of the main
ways in which they differ lies in the ability of the user running the test to configure the test
parameters. Silicon Graphic Inc.'s WebStone, the first benchmarking software specifically
developed to measure HTTP performance, is highly configurable. The idea is that the
results are more meaningful if they are tailored to a specific configuration. Why measure
retrieval time for a generic 10 KB HTML document when you can measure the time for the
actual document itself? The flip side of this argument is that standard tests better allow
administrators to analyze comparison data. Standard Performance Evaluation Corp.
(SPEC) calls this "apples-to-apples" comparison.
[Move on to the next part of the article]
Web Developer Site Feedback
Web Developer
Copyright © 1999-2001 internet.com Corp.
All rights reserved.
