This graph was created by munin. My linode is a 360. Is it normal how committed is way up there? If not, how can I diagnose the problem?

I have noticed that if I restart apache, the commited drops down to 1G, then climbs back up after a few minutes.

The Linux kernel overcommits it's memory. There's a lot of info out there about it, but here's some good links that explain it:

http://utcc.utoronto.ca/~cks/space/blog ... Overcommit
http://utcc.utoronto.ca/~cks/space/blog ... Overcommit
http://utcc.utoronto.ca/~cks/space/blog ... mitDefault

What that graph is telling you, is that if each process were to actually use what it says it needs, you'd be fubar

Take a look at your mysql setup too if you're running it on this server. Running <a href="http://blog.mysqltuner.com/">mysqltuner</a> will help you get that where it should be.

Hope that gets you started.

Justin

My vote goes for stack size.
In Linux, default stack size per-thread is 8MB. So, if you use, for example, apache-worker with many threads, or apache prefork with a bunch of subprocesses, the commit graph skyrockets.

I personally am running a 4x25-thread worker, and "ThreadStackSize 20972152" "freed"* me 600 MB of commit.
MySQL with InnoDB likes large overcommit too.

I think you can play with "ulimit -s" to cut down the stack size for apps that don't support configuring it explicitly, but keep in mind it may lead to badly-written apps crashing with out of memory errors.

*The memory wasn't actually ever allocated, it's just a flag somewhere in the kernel data structures.

I'm also running a apache instance with the worker MPM. Where exactly did you put that ThreadStackSize directive? In the apache2.conf under the "<IfModule mpm_worker_module>" section?

Yes, inside that section.
Also, this.