On Thu, 7 Jun 2012, Matt Clark wrote:
That's true, but if you look at slides 24 & 25 of
http://www.ddrdrive.com/zil_accelerator.pdf you'll see seek
distributions for a SLOG device with the ZILs for five filesystems -
although each ZIL is broadly append only (slides 21 and 22), there
is still a lot of seeking going on. Of course if you only had one
active FS on a pool, the activity would be almost pure append, so
sequential write performance would be the most important factor.
Agreed. I posted my response out of fear that someone might construe
your statement as meaning that the ZIL is also used to satisfy reads.
Bob
On 7 Jun 2012, at 00:10, Bob Friesenhahn wrote:
On Wed, 6 Jun 2012, Matt Clark wrote:
Yes, it's interesting to consider the possible real world performance. ZIL
writes, even by DDRDrive's stats, are very closely clustered around the tail of
the ZIL, with most seeks being to near locations - it's certainly not going to
seek more than 8 gig away. So probably just leaving it at one big partition
would be fine too.
The ZIL is not used for anything but to replay pending synchronous writes if
the system unexpectedly reboots. It is not used as a source of data while the
system is running. It is normally a write-only device.
_______________________________________________
OpenIndiana-discuss mailing list
[email protected]
http://openindiana.org/mailman/listinfo/openindiana-discuss
--
Bob Friesenhahn
[email protected], http://www.simplesystems.org/users/bfriesen/
GraphicsMagick Maintainer, http://www.GraphicsMagick.org/
_______________________________________________
OpenIndiana-discuss mailing list
[email protected]
http://openindiana.org/mailman/listinfo/openindiana-discuss
