Hey Andrew, Having a raidz3 vdev with two missing devices is significantly worse than raidz1. The computational costs would be far far higher. With raidz1 reading requires no reconstruction (and even then, raidz1 reconstruction is just xor). The raidz3 configuration you're describing would require expensive, 2-column reconstruction most of the time, and with a far more expensive mechanism than the simple xor of raidz1, doing calculations over a Galois field. For details:
https://github.com/illumos/illumos-gate/blob/b515258426fed6c7311fd3f1dea697cfbd4085c6/usr/src/uts/common/fs/zfs/vdev_raidz.c#L937 Adam .... and if this is someone from the ZFS developer summit trolling me, well done! ;-) On Mon, Nov 10, 2014 at 6:24 PM, Andrew Barnes <[email protected]> wrote: > Hi All, > > Hope you are all doing well, well done on the summit I have been keeping my > eye on the stream, many thanks. > > I've had a thought in the back of my head for a while... I assume this has > been asked before so apologies in advance. > > Is there any downside to a raidz3 with 2 missing devices v's a raidz1? > > If not, would it be viable to make ZFS always create raidz3's in the > background even if you ask for a raidz1 or a raidz2 and internally take into > account the raid level and adjust status reporting appropriately to allow > reshaping between the 3 levels? > > Many thanks, > > Andrew > > _______________________________________________ > developer mailing list > [email protected] > http://lists.open-zfs.org/mailman/listinfo/developer > -- Adam Leventhal CTO, Delphix http://blog.delphix.com/ahl _______________________________________________ developer mailing list [email protected] http://lists.open-zfs.org/mailman/listinfo/developer
