Macrobenchmarks

The first macrobenchmark consisted of compiling the Linux 2.4.2 kernel sources, since this involves significant file system activity and is representative of a workload experienced commonly by Linux users. The second macrobenchmark was the WebStone 2.5 benchmark for web servers [19], which is representative of a typical workload for a web server.

For the kernel compilation macrobenchmark, the time to execute ``make'' was measured. The 2.4.2 kernel sources were configured with the default options, and a ``make dep'' was done prior to the testing. Three kernel compilations were performed, each immediately after a reboot into single-user mode, and the results were averaged. This benchmark was executed on a 333MHz Pentium II with 128M RAM.

For the WebStone macrobenchmark, one hundred 10-minute trials were run with 32 web clients requesting the standard WebStone file set. A Sun Ultra 5 running SunOS 5.6 with 128M RAM was used as the test controller and client machine. This machine was directly connected using a 10Mbit Ethernet crossover cable to a 133MHz Pentium with 64M RAM running the Apache web server provided with RedHat 6.1.

 

Table: Macrobenchmark results. The elapsed and system times for a ``time make'' on the Linux 2.4.2 kernel sources are shown in minutes and seconds. The latency in seconds and throughput in MBits per second are shown for the WebStone benchmark.

	Base	SELinux	Overhead
elapsed	11:14	11:15	0%
system	00:49	00:51	4%
latency	0.56	0.56	0%
throughput	8.29	8.28	0%

 

Table 11 displays the results of the macrobenchmarks. There was no significant change in the total elapsed time, and there was only a 4% increase in the system time for a kernel compilation. There was no significant change in either the latency or the throughput measurements for WebStone. At the macro level, there appears to be little noticeable difference.