Rich Neswold <rich.nesw...@gmail.com> writes: > I just got a hold of a pair of Dell 2850s and, while building from source, > the system "froze" for several minutes. It eventually recovered. The logs > didn't show anything unusual (like out of buffers), so I didn't pursue it > aggressively. Doing a CVS checkout of PKGSRC also caused it, so it seems > related to very high, write traffic. These systems have AMI Hardware RAID > controllers (amr driver) and I figured when the write cache got full, it > prevented any other disk I/O until the writes finished. I rebooted and went > into the BIOS and shut off the write cache. I haven't seen the problem > since, but I became busy with other projects and haven't tried to diagnose > it further.
It seems you've got it. I turned off the delayed write functionality of the cache on my 2850, and the problem has gone away. To test it, I loaded the system down much more than previously, with various tasks, each of which has triggered the unwanted behaviour in the past: I ran a full build from source with "-j 4", and after observing for a while that that caused no problems, I added a full backup, a busy bittorrent client, and NFS service for another system build running on a client system. Still no hangs, and this sort of load would have ended with me hitting the power button to regain control over the box in the past. -tih -- It doesn't matter how beautiful your theory is, it doesn't matter how smart you are. If it doesn't agree with experiment, it's wrong. -Richard Feynman