Re: RaiserFS via NFS
On Tue, 2004-04-20 at 05:07, Markus Schabel wrote:
> Marcel Hicking wrote:
> > --Saturday, April 17, 2004 11:38:56 -0700 Chad Cranston <email@example.com>:
> >>I chose ext3 for it's reliablity over ReiserFS.
> > I found ext3 too slow (although rock solid) for large
> > amounts of mail. Since Reiser was no option (too much
> > data loss in the past) we opted for XFS.
> well, i see the same problem as everybody here: i've had some corrupted
> reiserfs systems, and it wasn't possible to restore the data (except
How long ago? Was this a recent kernel/reiserfs-tools?
> backups of coures ;)). We're still running reiserfs on our proxy servers
> (squid), but we have the phenomenon that the machines get slower and
> slower while squid is running, and if you stop squid and wait some time
> and start it again it all goes fine again. but the problem isn't squid,
> it seems to be reiserfs which seems to be not able get all data written
> to disk in time and slows the computer down. (sure this also depends on
> the harddisks, but we played around with hdparm and the situation was
> exactly the same with DMA enabled (140MB/s) and disabled (4MB/s), so it
> cannot be the HDD).
Sounds strange... have you mounted the squid partition with '-o
noatime'? This is a standard recommendation for squid as it reads lots
of little files and the updated atime writes can be a slowdown.
It sounds to me more like you have squid slowly running out of memory.
Check 'free' before and after you restart squid. Depending on how you
have configured your squid, it can start thrashing when the memory cache
fills up. It could be that reiser uses a little more memory, pushing
your squid over the edge.
I remember reading a report that tested Linux/squid on a heap of
different filesystems that concluded that the best was reiser... this
was some time ago though so things might have changed, and perhaps the
noatime was critical.
Donovan Baarda <firstname.lastname@example.org>