I''m pretty sure that this bug is fixed in Solaris 10U5, patch 127127-11 and 127128-11 (note: 6462690 sd driver should set SYNC_NV bit when issuing SYNCHRONIZE CACHE to SBC-2 devices). However, a test system with new 6140 arrays still seems to be suffering from lots of cache flushes. This is verified by setting "zfs_nocacheflush=1" and seeing noticeable improvement in performance. Can someone verify that this fix is indeed in the above release and patch set. If so, then are there parameters that need to be set for this to be active or is there firmware or other updates that need to be loaded into this new array? Any help appreciated. This message posted from opensolaris.org
Hello Bill,
Wednesday, June 4, 2008, 12:37:38 AM, you wrote:
BS> I''m pretty sure that this bug is fixed in Solaris 10U5, patch
BS> 127127-11 and 127128-11 (note: 6462690 sd driver should set
BS> SYNC_NV bit when issuing SYNCHRONIZE CACHE to SBC-2 devices).
BS> However, a test system with new 6140 arrays still seems to be
BS> suffering from lots of cache flushes. This is verified by setting
BS> "zfs_nocacheflush=1" and seeing noticeable improvement in
BS> performance. Can someone verify that this fix is indeed in the
BS> above release and patch set. If so, then are there parameters
BS> that need to be set for this to be active or is there firmware or
BS> other updates that need to be loaded into this new array? Any help
appreciated.
BS>
Well, the real question is how 6140 reacts to SYNC_NV - probably it
doesn''t care...
--
Best regards,
Robert mailto:milek at task.gda.pl
http://milek.blogspot.com
Hi Robert> > Well, the real question is how 6140 reacts to SYNC_NV > - probably it > doesn''t care... >That was our conclusion also, but its really hard to connect the dots... This message posted from opensolaris.org