If you are one of the poor people being forced to have gamin installed on your Linux-Box (e.g. in RedHat Enterprise Linux 4 you can't install gnome without gamin), you might notice that it sometimes creates a lot of CPU-usage and/or high load.
Read on for a workaround, some interesting hints about "load average", and a pretty diagram
We have a quite massive Linux-Server (2 Dual-Core Opterons, 16Gig of RAM), which is supposed to be a "login-server" for our users, meaning they connect to it graphically using X11 (to be exact: XDMCP) or NX and then start small programs.
After a few days of uptime, the load average started to go berserk: With 30 users which where basically doing nothing CPU-intensive, the load rose to 30, with peaks of 50. We started investigating, and I found something quite interesting: The "load average" is not an average! You can read about this in UNIXÂ® Load Average Part 1: How It Works and Part 2: Not Your Average Average
Some more investigation revealed that it was gamins fault. It was started for each user that used GNOME. It can't be killed (it automatically respawns), and neither can it be removed (too many packages depend on it). So after some tinkering we found a solution by disabling gamin for all filesystem-types, by putting the following into /etc/gamin/gaminrc: fsset nfs none
fsset reiserfs none
fsset ext3 none
fsset mvfs none
fsset smbfs none
fsset cifs none
fsset autofs none
Obviously this helped; here's the promised diagram, you can clearly see that we've "fixed" gamin at 12:18, and the exponential decrease of the load "average" afterwards: