responding to excerpts: On 6/16/2011 3:52 PM, Lawrence K. Chen, P.Eng. wrote: > Just to throw out my 2cents... > > We currently have a NetApp 3070? Was purchased quickly when we were having > problems with our NFS cluster (pair of v240's, sol9, vxvm+ufs from > 9980)....we eventually resolved the issue...there was a>1TB UFS filesystem > that had started out life as a 40MB filesystem and slowly grown over > time....and the users never delete anything on it....so every now then things > would just stop so it could search through the entire filesystem from > beginning to end for some free inodes. Even though somebody came in better > and lower to the bid, they pushed for NetApp and that's what we got. $.02, you should have spent a little more to get VxVM + VxFS. A defrag would have fixed you right up and the snapshots, journaling and speed are all great.
> Back when we ran our own email....we were running 2+TB ZFS filesystems for > mail spools (shared out over NFS to 3 MDA's and 8 imap/pop servers) The only > hiccup was we had to turn ZIL off, because ZFS insisting that our 9980 flush > cache and blocking everything until it did wasn't working too well (also ran > into an issue with IPF and NFS...it was a T2000 and IPF would pin one CPU and > cap our throughput....we knew we could do better from testing, but put 40,000 > users on it and couldn't get there...) For small files over NFS, zfs is sub-optimal when compared to UFS. a fast SSD ZIL makes the metadata performance rock for lots of small random IOs. For large file, sequential performance, you don't want a flash SSD ZIl because the controller to the ZIL becomes the bottleneck for throughput storage. YMMV. > Reportedly Solaris 11 does fix on problem we have with ZFS....right now it > doesn't tell anybody in logs or by email that its DEGRADED. We get emails > from VxVM when it loses a disk. I ended up making our cfengine check zpool > status, first time it ran....I discovered 5 degraded zpools. 2 were > dismissed as firmware bugs, so I had patch the drives.....a couple were where > a former admin had taken the disk out to do an upgrade and never put it back. > And, one was an actual bad disk (and Oracle surprised me by promptly > agreeing and sending me a new disk...) The 2 that were firmware (I inherited > the tickets)....it took 6 months to eventually get the downtime on those > production servers to patch the disks (at one point I was worried I had > bricked the drives on one of the servers, but turned out I had lost network) > I have a cron job that checks zpool output once a day to look for degraded disks. Sad that this is necessary in this day and age. It's one of the bits of enterprise polish that zfs lacks (along with the total inability to do a relayout, re-balance, and some other things). _______________________________________________ Tech mailing list [email protected] https://lists.lopsa.org/cgi-bin/mailman/listinfo/tech This list provided by the League of Professional System Administrators http://lopsa.org/
