Check out the new USENIX Web site. next up previous
Next: Test Environment Up: Experimental Methodology Previous: Experimental Methodology

Workload

We use two different synthetic workloads for our experiments: SPECWeb96 [26] and WebStone 2.5 [27]. SPECWeb96 was the first standard HTTP benchmark. The SPECWeb96 working set comprises files that range in size from 100 bytes to 900 kB, where small files are referenced more often than large files (50% of the total number of requests reference files smaller than 10 kB). In addition, the SPECWeb96 working set scales with the expected server throughput. In all of our experiments, the entire working set fit into the server's RAM, thus avoiding any performance distortion due to disk accesses.

SPECWeb96 has been superseded by SPECWeb99 as the industry-accepted Web serving benchmark. SPECWeb99 exercises HTTP/1.1 features, such as persistent connections, and includes requests for dynamically generated pages.

Although SPECWeb96 does not take into account some aspects of current HTTP workloads (e.g. no persistent connections, no dynamic content), it is well suited for measuring static file serving performance, which is the main purpose of our performance evaluation. Furthermore, large HTTP sites often use several servers that are partitioned into groups serving different types of content such as static files, user logins, and databases. The static content servers are likely to experience workloads similar to the SPECWeb96 workload. Finally, the SPECWeb96 execution guidelines are sufficiently strict as to allow meaningful comparison of independently reported results.

The results presented here do not meet SPECWeb96 reporting guidelines and are not certified SPECWeb96 results. The SPECWeb96 benchmark was executed for the largest workload corresponding to the reported result rather than ten evenly spaced lower throughput workloads as required by SPECWeb96 for reporting purposes. This does not affect the results reported in this paper.

WebStone is another HTTP server benchmark. Unlike SPECWeb96, it allows a user to change the workload characteristics, making it easier to identify performance bottlenecks for given file sizes. For WebStone, our workload consists of fixed-size files, ranging from 64 bytes to 1 MB. The file size is varied in each test.


next up previous
Next: Test Environment Up: Experimental Methodology Previous: Experimental Methodology
Philippe Joubert 2001-05-01