Hi,

I run a game hosting company and one of our nodes has recently shot up in I/O usage. kjournald uses anywhere between 92-99% IO usage as reported by iotop (2.6.39 kernel) on CentOS 5. I have absolutely no idea what to do with this but it's absolutely affecting the game server quality.

I've tried setting writeback on the filesystem (ext3) instead of ordered, however it's been to no avail. I understand that this could just be the overload on the disk, however the load is only 9 (there's 16 cores -- dual processor 8 cores each @ 2.4ghz with 48GB of DDR3 memory and the hard drive itself is a 15K SAS -- no raid), and we have another server running the exact same setup running double the amount of servers and players without a problem (kjournald stays at ~0.5%) and we even have SATA servers with kjournald staying at ~0.1-1% at roughly the same transfer amount, so I'm sure this is an issue with the server itself.

I get the same issue when reverting to 2.6.18 (no iotop support however everything still lags the same). Ideas?

First thing would be to fsck that filesystem. Then maybe think about noatime/nodiratime.

Another option would be to use an external journal and put it somewhere else - if it really is a journal issue.