On Tue, 2004-03-23 at 12:40, Bojan Smojver wrote: > I did two quick benchmarks with sendfile() support turned off in order > to determine rough impact of the patch on emulate_sendfile() and the > whole of Apache (prefork MPM).
Just did another set of runs on the same machine (this time 100,000 requests, 3 runs each; the file, 1456 bytes long, was specified in the URL, rather than being a directory index): Patched: 1,205.07 req/sec (average of 3 runs) Unpatched: 1,214.05 req/sec (average of 3 runs) The above equals 99.26% of the speed of unpatched or 0.74% speed loss of the patched. I guess that would vary depending on how one does the testing, but my bet is that we'd be losing 1% to 1.5% on small files in cases where sendfile() support doesn't exist or is turned off. On a big file (8,100,081 bytes long), the result over 1,000 requests (single run, just for a little comparison): Patched: 12.16 req/sec Unpatched: 12.04 req/sec The above (i.e. patched was faster than unpatched) is most likely due to some internal conditions on the machine (there was of course nothing else running, but the kernel might have been sorting out its bits or something) during the runs, but one would expect the patch to affect larger files less anyway. Anyway, hope this info helps in making the decision about the patch. -- Bojan
