[Beowulf] SATA vs SCSI drives

J. Andrew Rogers andrew at ceruleansystems.com
Mon Oct 11 11:08:09 PDT 2004

I purchase many, many terabytes of disk array every year, and use a
mixture of SCSI and (S)ATA depending on the specific application.  My
observations and experiences with the current crop of technology:

SCSI attached to a decent RAID controller (e.g. LSI MegaRAID) will
generally outperform a roughly equivalent SATA array for many purposes,
and if you have money to burn you can build significantly faster arrays.
 This is due to a combination of physically faster drives and mature
drive and controller implementations that work very well together.

That said, for single-process access, streaming, and similar, the
performance is largely similar.  A 10k SATA array will perform about as
well as a 10k SCSI array in most cases.  For applications that are bound
by access/seek times (e.g. databases), SCSI still seems to have
substantially more throughput in practice.  The bandwidth issue is
almost a non-issue in my experience, as you'll run into access/seek
limitations first for most apps.

So to summarize, they are mostly differentiated by the effective
access/seek throughput; SATA is the cheaper choice if you aren't
significantly bound by this parameter.  And as SATA firmware in both the
drives and controllers improves, and fast SCSI drive hardware is adapted
to SATA interfaces, I expect this gap to close.  It hasn't closed yet,
but in a couple years I expect it will be.

j. andrew rogers

More information about the Beowulf mailing list