[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Findings / Last Call for Help
- To: misc_(_at_)_openbsd_(_dot_)_org
- Subject: Re: Findings / Last Call for Help
- From: "Paul M. Hirsch" <elektrosatan_(_at_)_voltagenoir_(_dot_)_org>
- Date: Wed, 25 Apr 2001 02:13:10 -0500
On Tue, Apr 24, 2001 at 04:34:54PM -0700, rogue_(_at_)_ngxweb_(_dot_)_com wrote:
> OpenBSD 2.9 Beta (#613 or something like that) was about twice as slow as
> FBSD taking anywhere from 27-37 secs to change maps...Indeed, contributing
> to this slow time was the fact that it seemed that it would take anywhere
> 10-15 sec before OBSD would even begin loading...that is, it seemed to just
> site there for about 10-15 secs before it I would see the harddrive light
Hmmm... that 10-15 sec delay sounds like the real problem, and that
doesn't sound like a performance tuning issue (with the OS anyway.)
> In any event, I would appreciate any advice.
Have you tried running a trace on the server process?
# ktrace -ip <pid>; kdump -l
where <pid> is the PID of your HL server process. This will start a
system strace of the server process (and spawned children), sending
the trace to ./ktrace.out. kdump -l will read in the dump, and do
a tail -f style watch on it. (This is important so you can see
what system calls are being called when.)
Once you are tracing, try loading a new map, watching the dump
durring the 10-15 second delay. When the disk starts to fire
after the delay, you will want to ^C out quick, then you can
# ktrace -cp <pid>
To stop the trace dump.
Try to see what the server is waiting for durring the
delay. Hopefully something will turn up.
When it comes down to raw performance (IO, net, etc), OpenBSD is right
in there with FreeBSD, or very close behind, so I would be suprised if
your problem was caused by a performance limitation in OpenBSD.
/Paul M. Hirsch /
/GPGPGPkeyID: 0xD11A250E /
Visit your host, monkey.org