[nycbug-talk] Future Meeting: AMP on BSD

Charles Sprickman spork
Thu Aug 5 18:16:20 EDT 2004

On Thu, 5 Aug 2004, Hans Zaunere wrote:

> > I'd love to hear more about how to make mysql run right.  It's an
> ongoing
> > issue that comes up on the lists, and I for one still see issues under
> > moderate load with Linux Threads and the "skip-name-resolve" option.
> Interesting... which BSD?  I run MySQL both with native threads and
> LinuxThreads on FreeBSD without issue.  I'll certainly touch on some of
> this, and answer any questions people have.

FreeBSD 4.7, 4.8 and 4.9.  While going with LinuxThreads gets rid of the
"spinning at 99% cpu" problem, I've found that if the box is doing
anything else (especially mail delivery w/qmail or sendmail periodically
chugging through a queue), mysql just starts timing out on local (socket)
or network queries.  We're talking both select-only at 15 q/S tops here on
a database that only is about 20MB.  I recently brought this up on the
mysql list with lots of specifics (no, the machine's not swapping, disk io
is sane, cpu is not pegged).  The one thing that does seem to stand out is
that the box spends lots of time in "system" cpu and context switches as
qmail spikes, and this seems to be when either threading library starves
mysqld and the application (vpopmail) starts timing out.

> > Also some good tips on configuring php beyond the port build; what do
> you
> > recommend changing in the stock php.ini, what are some performance
> tips,
> > and which of the various accellerators do you find work best
> > (turck-mmcache looked good, but that project apparently died)?
> Good points, too  (hint: always use php.ini-recommended).

Excellent... thanks.  If you can live with Turck-mmcache's sig 11's on
HUPs, the speed increase on a single purpose application like squirrelmail
is staggering.

In general, I'm something of a stranger to php.ini and more specifically
getting a sane and usable safe-mode config going...  Hope you've got four
or five hours blocked out for the meeting. :)


> H

More information about the talk mailing list