On Sat, 2002-05-25 at 13:08, Brian Pane wrote:
> I'll run httpd benchmarks using this patch and the current pool code.
> If all goes well, I'll post results later today.
Good news:
The benchmark tests are finished, and I don't see any loss of
httpd performance with Sander's patch.
The test case:
httpd-2.0.37-dev on Solaris 8/Sparc
worker MPM, running on a server with 8 CPUs
no libmtmalloc
load generator sending 50 concurrent requests for a 0-byte file
over 100Mb/s switched LAN
The point of this test setup was to amplify the effect of any
extra mallocs that the patch might require.
Results:
Configuration Requests/sec CPU util Load avg
------------- ------------ -------- --------
2.0.37 baseline 1206 72% 7.3
plus Sander's patch 1213 71% 7.3
The difference in throughput and CPU utilization is within the
range of normal measurement error for this test setup, so I'd
say that the httpd performance is identical with and without
the patch.
--Brian
|