On Aug 4, 2009, at 1:35 PM, Bob Friesenhahn <bfrie...@simple.dallas.tx.us
> wrote:
On Tue, 4 Aug 2009, Ross Walker wrote:
But this MUST happen. If it doesn't then you are playing Russian
Roulette with your data, as a kernel panic can cause a loss of up to
1/8 of the size of your system's RAM (ZFS lazy write cache) of your
iSCSI target's data!
The actual risk (with recent zfs) seems to be 7/8ths RAM (not 1/8),
sufficient data to accomplish 5 seconds of 100% write, or up to 30
seconds of aggregation time. On a large memory system with high
performance I/O, this can represent a huge amount (gigabytes) of data.
Yikes! Worse then I thought.
Actually I recommend using a controller with an NVRAM cache on it,
say
256MB-512MB (or more).
This is much faster then SSD and has the advantage that the ZIL is
stripped across the pool making ZIL reads much faster!
Are you sure that it is faster than an SSD? The data is indeed
pushed closer to the disks, but there may be considerably more
latency associated with getting that data into the controller NVRAM
cache than there is into a dedicated slog SSD.
I don't see how, as the SSD is behind a controller it still must make
it to the controller.
Remember that only synchronous writes go into the slog but all
writes must pass through the controller's NVRAM, and so synchronous
writes may need to wait for other I/Os to make it to controller
NVRAM cache before their turn comes. There may also be read
requests queued in the same I/O channel which are queued before the
synchronous write request.
Well the duplexing benefit you mention does hold true. That's a
complex real-world scenario that would be hard to benchmark in
production.
Tests done by others show a considerable NFS write speed advantage
when using a dedicated slog SSD rather than a controller's NVRAM
cache.
I get pretty good NFS write speeds with NVRAM (40MB/s 4k sequential
write). It's a Dell PERC 6/e with 512MB onboard.
The slog SSD is a dedicated function device so there is minimal
access latency.
There is still bus and controller plus SSD latency. I suppose one
could use a pair of disks as an slog mirror, enable NVRAM just for
those and let the others do write-through with their disk caches
enabled and there, dedicated slog device with NVRAM speed. It would be
even better to have a pair of SSDs behind the NVRAM, but it's hard to
find compatible SSDs for these controllers, Dell currently doesn't
even support SSDs in their RAID products :-(
-Ross
_______________________________________________
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
