WebStone 2.0 is the second generation Webstone web server benchmark. It incorporates numerous bug fixes, modifications for compatibility with other platforms and adds the new functionality of benchmark proxy servers, cgi and NSAPI programs as well as introducing run rules which should make Webstone numbers significantly more meaningful for comparison.
Webstone 2.0 provides facilities for benchmarking proxy servers. This is accomplished by putting in a value for the the PROXYSERVER entry in the conf/testbed file, and changing the filelist to include URL's that have the hostname for the actual web server.
Dynamic content benchmarking is now explicitly supported in Webstone 2.0. The file README.DynamicWorkload has directions for testing of NSAPI. The included filelist.dynamic-{light,medium,heavy} serve as sample loads, with the filelist.dynamic-heavy being the load that should be reported for NSAPI performance. The cgi-send numbers should be quored for the filelist.cgi-heavy fileset.
A port of the WebStone 2.0 benchmark to Windows NT is also included in this release. This port is still in progress, so full functionality is not assured. Specifically only the benchmark code has been ported - the supporting scripts have not.
As of Webstone 2.0, there are now run rules which must be adhered to for published Webstone numbers. These are fairly basic, but they provide important constraints on the benchmarking which make Webstone numbers more meaningful.
Fileset: Included in the Webstone distribution is filelist.standard, which was previously called filelist.sample. This filelist has a distribution of fileset sizes that matches the kind of distributions seen in live web sites. The largest file in the distribution is a 5 MB in length, which simulates the occasional MPEG or other animation file which is downloaded. This filelist should be used for all published Webstone numbers. Note that running WebStone 2.0 with the sort of fileset given in WebStone 1.1 will not yield a comparable benchmark. In general, the WebStone 2.0 filelist will yield lower rates for connections/second, but higher rates for throughput - the two sets of numbers cannot be compared.
When reporting NSAPI numbers, the filelist.dynamic-heavy filelist should be used. For CGI numbers, the filelist.cgi-heavy filelist should be used.
Benchmark Run Configuration: For a reported WebStone run, the runtime must be set at least 10 minutes. This provides adequate time for the server and client configuration to reach a steady state, and then provides a length of time long enough to cancel out the high variations seen in the first few minutes of the run. The number of clients should also vary from 20 to 100 in increments of 10 so that performance of the server under a wide variety of loads can be observed.
Server Configuration: The number of threads/processes is open to the discretion of the benchmarkers. However, whether server side logging is on must be explicitly reported. When logging is turned on, it must be in the common logfile format, and only IP addresses should be logged. Parsed HTML is recommended to be turned off.
Proxy Configuration: The configuration of how often the proxy server polls the actual server for refreshes of it's cache should be described, as well as any information
Server Machine Configuration: When reporting runs, it is necessary that the operating system, memory configuration and any special operating system modifications be reported, especially changes to the TCP/IP stack.
Testbed configuration: Reported runs must include information about the network topology being used, as well as the number and type of machines generating load.
All reported runs must include the information summarized by webstone -results, excluding the timestamp. This includes: number of clients, connections per second, little's law number, latency, error level and throughput. Preferably in a table format.