[The sounds of tristan rapidly googling]
Ack, you're right, it does look like zfs will spread reads over both
drives. Oops. For those reading my last post - sorry - I'm not sure
where I picked up the idea that the zfs reads from both sides of the mirror.
Do you know how close to the data the zfs checksum is stored? Is it
possible that one FS block read will actually result in two seperate
physical IO's, one for the data, one for the checksum?
Roman, rather than reading my mis-interpretations, perhaps check:
http://blogs.sun.com/roch/entry/when_to_and_not_to instead - it explains
everything I was trying to say, except far more accurately. :-)
T
Bob Friesenhahn wrote:
On Thu, 20 Aug 2009, Tristan Ball wrote:
I believe you will actually get the read IOP's rate of 4 disks,
rather than 8. ZFS doesn't load balance reads accross mirrrors like
raid does, rather it reads from both mirrors so it can compare
checksums.
Why would zfs want to read from both mirror devices so that it can
compare checksums? That is useless work which does not make any sense
at all.
It may be that zfs does not carefully balance reads across mirrors but
it does do non-preferential load-balanced reads using either device in
the pair. Statistically this results in measurable benefit. Someday
maybe zfs will become smarter and carefully direct the I/O requests
for more benefit with mirrors.
But happy to be shown wrong if someone can provider references? :-)
Smoking Mirrors!
Bob
--
Bob Friesenhahn
[email protected],
http://www.simplesystems.org/users/bfriesen/
GraphicsMagick Maintainer, http://www.GraphicsMagick.org/
______________________________________________________________________
This email has been scanned by the MessageLabs Email Security System.
For more information please visit http://www.messagelabs.com/email
______________________________________________________________________
_______________________________________________
storage-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/storage-discuss
