[wplug] RAID performance

[Brandon Poyner]

On 3/2/07, Mike S <lists at immuneit.com> wrote:

> However, the thing that sticks out in my mind about your test is the difference in drives --
> SATA I vs SATA II.  My impression was that SATA II was a big improvement over SATA I.

Yes, you can see significant improvements with SATA II if the drive
heads have a lot of seeking to do.  SATA II has native command queuing
(NCQ) which is similar to SCSI tagged queuing, and if the
motherboard's chipset also supports NCQ you can see iowait times drop
significantly.

Also be careful with generic benchmarking.  For instance there are two
things wrong with "time tar --totals -cvpf /dev/null /directory".  The
first problem is that the file can be cached and the second iteration
of tar will might end up using the cache.   The second problem is that
gnutar has a "feature" to detect that stdout is going to /dev/null and
that you "only" want the size estimate and will *not* actually read
the file contents.

-- 
Brandon
Kiva.org - Make a small loan, Make a big difference