[imp] system / imp tuning

Eric Rostetter eric.rostetter@physics.utexas.edu
Wed Nov 6 20:22:20 2002


Quoting Colin Devine <devinec@sas.upenn.edu>:

> Hello,
> 
> We recently upgraded our Horde/Imp install from horde 1.2.6/imp 2.2.7 to
> 2.1/3.1.  Also included in this was an upgrade to php-4.1.2.  Since then
> we notice that our apache processes grow in the amount of memory they
> use as time progresses, slowing down the server and IMP itself.  At the
> start of the apache server they are ~7M but in about 24 hours they are
> often ~25M in size.  This seems to me to indicate a memory leak of some
> sort and I was wondering what the collective knowledge base of this list
> had any ideas.

That would seem to be a real problem, but I don't know what exactly.
My processes never grow over 8-9M.  Of course, how large they are depends
on your php installation (the more you crame in php, the more memory the
module will take).

> I include my ideas as starting points.
> 
> 1) Restart the server every x hours, where x is at least 24.  We did
> this previously with the old horde/imp and perhaps this was helping us
> more than we realized.

This may provide temporary relief, but it doesn't fix the problem.  You 
need to find and fix the real problem.

> 2) Change MaxRequestsPerChild to maybe 5000-10000 instead of 0 (0 being
> the child processes never die).

Probably a good idea, but again it won't fix the problem.  You should
still find and fix the real problem.

> We are also investigating php accelerators, imap proxies, and even
> multiple clustered webmail servers in an effort to load balance and
> generally speed up IMP.

All valid ideas, but you should really fix this php/httpd problem first.
The accelerators won't help if you are out of memory, and throwing more
machines at it is the expensive way to fix a memory problem.

> All thoughts comments and questions are encouraged.

Well, I have no problem with most of your ideas, except that they seem to be
to avoid fixing a problem.  I wouldn't bother to restart the server, but
all the others are good ideas.

Maybe someone else on the list will be able to help with the memory issue,
but my guess is you really want to ask the php and/or httpd mailing list
about this, not the IMP list.

-- 
Eric Rostetter
The Department of Physics
The University of Texas at Austin

Why get even? Get odd!


More information about the imp mailing list