Didn't make it to the list, so resending from another account. On Fri, 2006-07-28 at 13:31 +0200, Goswin von Brederlow wrote: > > We (at my workplace) have lots of them, smp and not, with reiserfs and > they don't usualy crash. They do crash a lot when we get new ones > untill we weed out all the bad ram and such but after that the > majority runs stable. The rest we swap cpu or the mainboard till they > work. > > We still do have problems with reiserfs every now and then > though. Having power getting cut from nodes without proper shutdown > seems to be a problem for reiserfs. On reboot the syslogd hangs for > ages unless /var is reformated. Recently I convinced my boss to switch > to another filesystem but we still have to test crash (e.g. pull the > power every 5 minutes) the different FSes a lot to see which is most > robust. > > Personaly I use ext3 and never had problems on amd64. > I'm quite surprised to hear all these bad opinions about reiserfs since I have never had any problems with it not caused by bad hardware.
One thing that many people seem to be missing is the fact that their drives might have write-cache enabled. In that case, even journalling filesystems can be damaged by non-clean unmount if they don't handle the caching issues. Meta-data corruption in ext3 might be less visible because of the defensive layout of its structures. You can use hdparm/sdparm to find out if the write-cache is enabled on your drives and disable it. Note that disabling write-cache can cause performance loss and decrease MTBF of the drive. If you use reiserfs or ext3 then you can keep write-cache enabled, since those filesystems support that setup via mount options. Use barrier=flush for reiserfs and barrier=1 for ext3. See this website for more info: http://www-dt.e-technik.uni-dortmund.de/~ma/linux/kernel/safe-write-caches.html Michal Palka ---------------------------------------------------------------------- PS. Fajny portal... >>> http://link.interia.pl/f196a -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]