httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Martin Ramshaw" <>
Subject Re: [PATCH] Solaris sockets for support/ab.c (httpd-2_0_32-beta)
Date Wed, 13 Mar 2002 22:33:41 GMT

>I hope this helps understand the big picture (at least as I see it).

It does, and thank you for taking the time to explain this to me.

My hope was to be able to foist off benchmarking onto someone
else, which isn't really an option if too much configuration (i.e.
is required.

My frustration was that (as you confirmed) a benchmarking effort
involves benchmarking the machine driving the benchmark (i.e.
running 'ab') as much as anything else. Of course, this is less of an
issue now that Solaris Intel is to be mothballed.

Unfortunately, benchmarking (MTBF, Average and distributed
response times, whether failure is graceful or not, suggested hardware
platform for desired performance, etc, etc) quickly turns into a study
in comparative operating systems.

NT starts rejecting connections at around 58 concurrent connections;
it turns out that NT is not a *true* multi-threaded OS at all but has an
internal limit of 64 concurrencies, some of which it needs itself.

My experience with other OS's is that they gracefully start rejecting
connections at a much higher level of concurrency.

    Best regards

View raw message