OpenSolaris 2008.11 ZFS Performance

Computing Science, posix rules, life rules, no rules…: Publications; Comments

Marcelo Leal on December 30, 2008 — 6 Comments

A simple test to see how the ZIL code behaves on a configuration where the slog could be a bottleneck…
ZPOOL with 12 disks(750gb each) in load share:

ZFS

ZPOOL with 12 disks(750gb each) in load share, plus a 32gb slog:

ZFS

The ZIL code should not avoid write to the slog if the pool have a better throughput?

Share and Enjoy:

You can follow all the replies to this entry through the comments feed

andrewk8

December 30, 2008 - 5:05 pm | Permalink

It doesn’t really make sense to allocate a dedicated slog if it is slower than the array. Eventually ZFS should hopefully cope with this situation, but I still consider it a work in progress, impressive though it is.
Marcelo Leal

December 30, 2008 - 5:20 pm | Permalink

Yes, but that is why i did the test.
In this scenario it can be easy to see, but the situation can be a bit more complex than that. In that case, we will need the ZIL super power. ;-)
SSD is in vogue, some people could think it is the solution for all configurations.
Marcelo Leal

January 5, 2009 - 10:29 am | Permalink

Talking with Roch Bourbonnais, the fix is in development:
http://bugs.opensolaris.org/bugdatabase/view_bug.do?bug_id=6706578

Leal
carton

January 8, 2009 - 7:44 pm | Permalink

Roch says there is a fix already in the Storage 7000 array, but not in the open source release:

http://mail.opensolaris.org/pipermail/zfs-discuss/2009-January/054148.html

blog