On Wed, Jun 24 at 18:43, Bob Friesenhahn wrote:
On Wed, 24 Jun 2009, Eric D. Mudama wrote:
The main purpose for using SSDs with ZFS is to reduce latencies for
synchronous writes required by network file service and databases.
In the "available 5 months ago" category, the Intel X25-E will write
sequentially at ~170MB/s according to the datasheets. That is faster
than most, if not all rotating media today.
Sounds good. Is that is after the whole device has been re-written a
few times or just when you first use it?
Based on the various review sites, some tests experience a temporary
performance decrease when performing sequential IO over the top of
previously randomly written data, which resolves in some short time
period.
I am not convinced that simply writing the devices makes them slower.
Actual performance will be workload specific, YMMV.
How many of these devices do you own and use?
I own two of them personally, and work with many every day.
Seagate Cheetah drives can now support a sustained data rate of
204MB/second. That is with 600GB capacity rather than 64GB and at a
similar price point (i.e. 10X less cost per GB). Or you can just
RAID-0 a few cheaper rotating rust drives and achieve a huge
sequential data rate.
True. In $ per sequential GB/s, rotating rust still wins by far.
However, your comment about all flash being slower than rotating at
sequential writes was mistaken. Even at 10x the price, if you're
working with a dataset that needs random IO, the $ per IOP from flash
can be significantly greater than any amount of rust, and typically
with much lower power consumption to boot.
Obviously the primary benefits of SSDs aren't in sequential
reads/writes, but they're not necessarilly complete dogs there either.
--eric
--
Eric D. Mudama
edmud...@mail.bounceswoosh.org
_______________________________________________
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
