Hi Marek,
> Our company is in the process of moving the production server data on a
> enterprise storage servers and therefore we are considering which setup
> would have the best performance and stability at the same time. Our
> preliminary tests showed us that, at the filesystem level, XFS performs
> best. Unfortunately, to be capable of growing the volumes we need to add
> another layer - the volume manager. This increases the chances for
> something to go wrong.
> As I've read through the mailing list and other sources, there are only two
> possible software packages on Linux - LVM and EVMS. From what I've read I
> understood that LVM 1.x is feature-frozen and maybe even bugsolve-frozen in
> favor of LVM2 development (see the deadlock while using xfs_freeze issue).
> LVM 2.x is unstable beta and EVMS is so fresh that no real experience has
> been heard of.
we have made up a similar setup here.
Basically it's LVM stacked on top of soft RAID 5 and is in production
since June last year, with the caution of keeping away from snapshots (i've
had problems at the very beginning with Oops caused by the two stacked layers
and we can avoid snapshots at the moment) and making an external log (on
raid 1 preferably) for performance. We are currently stick to 2.4.18
(XFS 1.1 and LVM shipped with). In the meanwhile we have performed
various operations (creating, deleting, resizing (growing)) on volumes,
experienced two IBM DTLA disks failure (giving us the time to replace
things ;)) and all without a single problem at filesystem level.
We use AMANDA as backup software and the only problem we experienced was
on Oops rising from a bad interaction between xfsdump and NFS, now solved
(thanks Eric and Steve).
This fileserver exports via NFS, SAMBA and netatalk.
I've not tried EVMS, but it seems very promising.
> I understand that there's no bug-free software and no guarantee of anything
> but maybe some of you have some enterprise-level experience with, say, 2+
> TB data pool on XFS and some volume manager. I appreciate any reference of
> any kind, some known imcompatibilies, what mistakes to avoid.. it would
> help me a lot.
Our storage size is an order of magnitude less, so YMMV.
Ciao,
-m
|