Hi Prashanth, This was about a year ago. I believe I ran bonnie++ and IOzone tests. Tried also to simulate an OLTP load. The 15-20% overhead for ZFS was vs. UFS on a raw disk...UFS on SVM was almost exactly 15% lower performance than raw UFS. UFS and XFS on raw disk were pretty similar in terms of performance, until you got into small files...then XFS bogged down really badly. None of this was testing with snapshots, so I'm not sure of the effect there.
I can attest we're running ZFS right now in production on a Thumper serving two MySQL instances, under an 80/20 write/read load. We use ZFS snapshots as our primary backup mechanism (flush/lock the tables, flush the logs, snap, release the locks). At the moment we have 60 ZFS snapshots across 4 filesystems (one FS per zpool). Our primary database zpool has 26 of those snapshots, and the primary DB log zpool has another 26 snapshots. Overall, we haven't noticed any performance degradation in our database serving performance. I don't have hard benchmark numbers for you on this, but anecdotally it works very well. There have been some folks complaining here of snapshot numbers in the 200+ range causing performance problems on a single FS. We don't plan to have more than about 40 snapshots on an FS right now. Hope this is somewhat helpful. Its been a long time (2+ years) since I've used Ext3 on a Linux system, so I couldn't give you a comparative benchmark. Good luck! :-) Best Regards, Jason On 1/23/07, Prashanth Radhakrishnan <[EMAIL PROTECTED]> wrote:
Hi Jason, > My company did a lot of LVM+XFS vs. SVM+UFS testing in addition to > ZFS. Overall, LVM's overhead is abysmal. We witnessed performance hits > of 50%+. SVM only reduced performance by about 15%. ZFS was similar, > though a tad higher. Yes, LVM snapshots' overhead is high. But I've seen that as you start increasing the chunksize, they get better (though, with higher space usage). So, you saw performance reductions as much as 15% with ZFS clones/snapshots. I'm curious to know what tests and ZFS config (# of snapshots/clones) you ran on. I ran bonnie++ and din't notice any perceptible drops in the numbers. Though my config had only upto 3 clones and 3 snapshots for each of them. > Also, my understanding is you can't write to a ZFS snapshot...unless > you clone it. Perhaps, someone who knows more than I can clarify. Right. I wanted to check if creating snapshots affected the performance of the origin FS/clone. Thanks, Prashanth > On 1/23/07, Prashanth Radhakrishnan <[EMAIL PROTECTED]> wrote: > > > > > > Is there someway to synchronously mount a ZFS filesystem? > > > > '-o sync' does not appear to be honoured. > > > > > > No there isn't. Why do you think it is necessary? > > > > Specifically, I was trying to compare ZFS snapshots with LVM snapshots on > > Linux. One of the tests does writes to an ext3FS (that's on top of an LVM > > snapshot) mounted synchronously, in order to measure the real > > Copy-on-write overhead. So, I was wondering if I could do the same with > > ZFS. Seems not. > > > > Thanks. > > _______________________________________________ > > zfs-discuss mailing list > > zfs-discuss@opensolaris.org > > http://mail.opensolaris.org/mailman/listinfo/zfs-discuss > > > >
_______________________________________________ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
