Andrey Kuzmin wrote:
> As to X25-E (released today), this indeed would be a worthwhile try as
> both ZIL and L2ARC device (I believe the latter, if not both, has
> already been explored at Sun).

Update to my saga:  I'm thoroughly disappointed with Mtron 3035 drives, but
maybe they are just not suited for server use at all.  The random write
performance is worse than a mechanical drive!

At this point, we're looking to put ZIL on an NVRAM-based device, the Gigabyte
I-RAM with 4GB of memory (its max.)
http://www.gigabyte.com.tw/Products/Storage/Products_Overview.aspx?ProductID=2180

In advance of getting that, I did a rough approximation of its performance by
creating a 4GB file in /tmp and lofi-mounting it.  I turned off swap to ensure
the file didn't ever get swapped out of main memory (I have 32GB in this box, so
it's not an issue, and it's a test system anyway.)

# zpool destroy data
# swap -d /dev/zvol/dsk/rpool/swap
# mkfile 4g /tmp/slogfile
# lofiadm -a /tmp/slogfile
# zpool create data raidz \
    c0t2d0 c0t3d0 c0t4d0 c0t5d0 c0t6d0 c0t7d0 log /dev/lofi/1

Of course, I can't reboot or crash without losing my ZIL, but this is a test
box, and I'd never run it this way in production.  Nevertheless, performance is
impressive, as we expect:
                    extended device statistics
    r/s    w/s   kr/s   kw/s wait actv wsvc_t asvc_t  %w  %b device
    0.0 6148.2    0.0 786974.6 26.6  3.4    4.3    0.6  86  86 lofi1

>6000 write IOPS, or roughly 200x more than we were getting with the Mtrons.

We'll also be evaluating the Intel X-25E as soon as we can get our grubby little
hands on one.  :)

Eric
_______________________________________________
storage-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/storage-discuss

Reply via email to