On Nov 11, 2010, at 15:08, Kyle McDonald wrote:
> Any opinions? stories? other models I missed?
>
> Other questions:
>
> 1) The ZIL will be small compared to the size of these, can I use
> the rest as L2ARC or is that not such a good idea?
>
> 2) Will ZFS align the ZIL writes in such a way that those IOPS
> numbers will be close to attainable?
All the models listed are seem to be based on the Sandforce SF-1200.
There was some discussion on Sandforce''s technology on the list a
little while back, and to me the general consensus was that there were
concerns about how caching was done and what would occur if the power
was suddenly disconnected:
http://mail.opensolaris.org/pipermail/zfs-discuss/2010-May/thread.html#41326
http://mail.opensolaris.org/pipermail/zfs-discuss/2010-May/thread.html#41588
http://mail.opensolaris.org/pipermail/zfs-discuss/2010-June/thread.html#42298
Since you were looking at NewEgg, you may want to look at the OCZ
Vertex 2 Pro:
http://www.newegg.com/Product/Product.aspx?Item=N82E16820227534
It''s based on the SF-1500 chip set, has a supercap in case the power
goes out, and ECC of some kind (perhaps for the DRAM cache?). You''re
getting less storage for the dollar, but it should still be more than
sufficient for a "slog" device, and be more resilient.
To answer your questions:
1. Yes, that is technically possible. What''s "best" is
probably
determined by benchmarking / testing things out with the workload
you''re expecting to have.
2. It could: depends on the partition is properly aligned and/or
whether the disk itself can be queried for sector size (and if it lies
about it).
For (2), see:
a. PSARC/2008/769: Multiple disk sector size support
http://arc.opensolaris.org/caselog/PSARC/2008/769/
b. PSARC/2010/296: Add tunable to control RMW for Flash Devices
http://arc.opensolaris.org/caselog/PSARC/2010/296/
(a) appears to have been fixed in snv_118 or so:
http://bugs.opensolaris.org/bugdatabase/view_bug.do?bug_id=6710930
Not sure if ZFS takes advantage of this information, or it it''s only
the lower-level I/O code.
Also ran across the "ashift" property of zpools:
http://tinyurl.com/25gmy7o
http://www.solarismen.de/archives/5-Solaris-and-the-new-4K-Sector-Disks-e.g.-WDxxEARS-Part-2.html
There''s been quite some discussion on the topic over the last little
while:
http://www.google.com/search?q=zfs+4k+sectors