On Mon, 11 Aug 2008, Martin Svensson wrote: > > Granted, the simple striped configuration is fast, and of course > with no redundancy. But I don't understand how a mirrored > configuration can perform as good when you sacrifice half of your > disks for redundancy. Doesn't a mirror perform as one device? Can > someone please clarify the example from the above, I think I am > missing something?
There are a few wonderful things about the simple zfs mirror vdev. One of those really wonderful things is that a read can be satisfied by either device in the mirror and zfs will nicely schedule sequential reads so that both devices are used. This almost doubles sequential read performance. Similarly, for random read I/O, the mirrors provide the same number of spindles as a non-redundant stripe would. So you get more IOPS when reading. Zfs is also quite effective when writing data to a mirror pair. I don't know what zfs's internal rules are, but there does not appear to be much actual overhead from writing to a mirror. Zfs can decouple writes to devices in a mirror pair so that one device is allowed to get ahead of the other since it knows how to reconcile the differences in case the system crashes or the device stops responding. This is virtually impossible to do with hardware RAID-1 (other than some possible NVRAM caching) since the two disks are truely mirrors of each other. The main cost of zfs mirrors is the raw disk space consumed. If raw disk space is more important to you than performance, then consider using raidz2 since it makes maximal use of disk as well as offering more data security. Raidz and raidz2 waste precious IOPS whereas mirrors waste less-precious raw disk space. Bob ====================================== Bob Friesenhahn [EMAIL PROTECTED], http://www.simplesystems.org/users/bfriesen/ GraphicsMagick Maintainer, http://www.GraphicsMagick.org/ _______________________________________________ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
