Date: Tue, 10 Jul 2001 10:00:39 -0700 (PDT) From: Matt Dillon <dillon@earth.backplane.com> To: Michal Mertl <mime@traveller.cz> Cc: freebsd-smp@FreeBSD.ORG Subject: Re: Is SMP BETA in STABLE? Message-ID: <200107101700.f6AH0dQ13994@earth.backplane.com> References: <Pine.BSO.4.21.0107100847070.14739-100000@prg.traveller.cz>
next in thread | previous in thread | raw e-mail | index | archive | help
:> just copying bits, it will be running mostly in the kernel. If your
:> web server is doing significant userland processing, you will see both
:> cpu's in use - that is assuming that you have a significant enough
:> web load.
:>
:Thanks. That clarifies the thing.
:
:The consequence of this is that as I assume a lot of web server processing
:is quite simple this 'one kernel at a time' seems quite limiting for
:serving apache on stable. Do you know if the people telling SMP in Linux
:kernel (say 2.2 and 2.4) is better (I mean better performing)
:right? I won't switch to Linux anyway :-).
:
:The page I tested on was in perl - I bechmarked once with mod-perl and
:once without (got interesting results - on dual p3-600 with mod-perl I got
:17 pages/second and without 1.6) and resulting html was about 1.5Kb. Both
:cases showed a little more than 50% cpu utilization. This could mean than
:(even with a bit more complicated page) the overhead of connection setup
:(may be mostly Apache's fault) and data transfer (even on loopback) is
:crucial.
:
:I'll do some more benchmarking because I thing it's not correct setup to
:benchmark on localhost. I'll also look at ApacheBench. It's implementation
:may quite easily affect the results too.
:
:This is getting quite off-topic, but I think that lot of people interested
:in FreeBSD on SMP machines do on their servers webserving so they might be
:a bit interested.
:
:--
:Michal Mertl
:mime@traveller.cz
Well, that sounds to me like the benchmark is flawed. Apache and
FreeBSD kernel doing nothing but accepting connections and pushing
data can do hundreds of connections per second. It sounds like your
benchmark is serializing its requests to the web server, running just
one at a time, resulting in only one [mod]perl running at any given
moment, which of course would only be able to utilize a single cpu
(one non-threaded process can only run on one cpu in any SMP
environment). If it's perl it is probably cpu bound. I would imagine
that with a proper test you would find both cpu's in full use.
-Matt
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-smp" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200107101700.f6AH0dQ13994>