We performed our experiments on PCs with a 700 MHz Pentium III running FreeBSD 4.1 with an 18 GB IBM 10,000 RPM Ultra2 LVD SCSI (Ultrastar 18LZX). We also performed experiments on an SGI Origin 2000 with 18 GB 10,000 RPM Seagate Cheetah disks (ST118202FC) under IRIX 6.5. The results with different parameter settings were similar to each other so we only present a representative subset of the results.
Our experiments used the 4-day web proxy log as input into the workload generators. Among other measurements, we measured the proxy hit rate, which represents how frequently the web page will not have to be fetched from the server, and the file system read time, which represents how long the proxy must wait to get the file from the file system. The file system write time is the time it takes the file system to return after a write; this may not include the time to write the file or file meta-data to disk. (We call the file system read (write) times FS read (write) time in our figures and tables.) wg-Squid and Hummingbird measured the file system read/write times directly, and the output of the iostat -o and sar commands were used to determine the disk I/O times. We also report the throughput, which we compute as the ratio of the experiment run time and the total number of requests processed. The measurements are averaged over the entire log; warm-up effects were insignificant.