[K12OSN] SATA vs. SCSI

[Ken Meyer]

A couple of factors to consider.  Go to:

	http://www.storagereview.com

Scroll WAY down the long home page and check-out the articles there.  The
real comparison results may be somewhat difficult to wade through, but I
think are worth the time to try to understand which benchmark is applicable
to what.

The major points that I personally derive on these issues are:

1) RAID 0 is a very bad deal.  It does not provide very much performance
increase (apparently in part due to the caching algorithms embedded in the
typical drive's firmware), AND you have doubled your jeopardy because the
failure of EITHER drive will eat your lunch (See Comment One below, copied
from the S/R site untouched by human-driven keyboards).

2) The superiority of SCSI, given similar physical characteristics (RPM, bit
density, et al) of the drives, has been most evident in servers having
multi-user, random access to many small files, and that this advantage has
been due to "elevator queuing" in which the drive actively manages the order
of retrieving data.  ATA drives never implemented this feature because the
advantage in single-user system was small or even negative.  However, SATA
drives are encroaching on this capability, called TCQ (Tagged Command
Queuing) by WD and generically, and NCQ (Native...) by Maxtor and Seagate.

It seems, from the results, that the SATA people doing this are getting the
range on SCSI in server applications, and at significantly reduced cost (See
Comment Two below). But again, this is only really effective on servers (and
I think, best on servers coughing-up many small files from random
locations).  This is newish technology, so caveat emptor applies as usual.
Note that a controller card is required to put the data back on the PCI bus.
Not sure whether this applies to non-RAID configurations, though.

3) A question.  It has been my understanding that writing data to, say, a
RAID 5 array was SLOWER than to a single drive, since space had to be
allocated for pieces to be put on each of the drives, but that reading might
be faster (ATA or SCSI, regardless of TCQ or not).  So, does one get better
performance by putting all drives in an independent configuration, maybe as
a single volume, than as a RAID array -- reliability and the "redundancy
tax" aside?

Also note that it appears that we are finally getting to the situation where
the transfer rate choke-point may be the traditional PCI bus itself.

------>> Comment One <<--------------------

Results culled from RAID0 arrays as drive counts increase are interesting
from an academic standpoint, mainly due to the linear nature in which one
can add independent actuators. Admittedly, however, practical use of RAID0
in production servers is quite limited -- performance gains are more than
offset by the significant increase in risk. Should one fail in a four-drive
striped array, all data would be lost.

------>> Comment Two <<-------------------

A quick check at the time of this writing with StorageReview sponsor,
HyperMicro, prices 73 GB Cheetah 10K.6's at $339 each, Raptor WD740GD's at
$219 each, and the AcceleRaid 170 at $379. When released this August, the
Promise TX4200's price will be on par with that of the S150 TX4 that it is
meant to displace. It runs $159 at HyperMicro.

Hence, the following pricing (excluding cables and accessories) arises:

$1356 ($339 x 4 Cheetahs) + $379 (AcceleRaid 170) = $1735 (4-drive SCSI RAID
Array)

$876 ($219 x 4 Raptors) + $159 (FastTrak TX4200) = $1035 (4-drive SATA RAID
Array)

The cost difference between the two arrays works out to 40% -- significant
indeed. What does one sacrifice by opting for the less costly SATA solution?
After all, it has been demonstrated that SATA's performance is competitive
with, and in some cases exceeds, that of a comparable SCSI solution.

While WD has delivered a solution that can match a SCSI-based solution's
speed and scalability, one must also keep in mind the key factors of
infrastructure and reliability. As with TCQ itself, SATA's support hardware
such as back-planes, all-in-one solutions, and the like remain in their
infancy when contrasted to the maturity and longevity of SCSI hardware.

Also keep in mind that, while Western Digital claims an enterprise-class 1.2
million hour MTTF spec and backs the Raptors with a 5-year warranty, the
line is still new and remains relatively unproven compared to established
solutions such as Seagate's Cheetah series. Finally, remember that the
prices listed above represent the cost of the storage subsystem alone --
factoring in the total cost of server hardware when motherboards, CPU, and
RAM are considered can dilute the difference significantly.

In the end, the potential for SATA to invade the entry- and mid-level server
market is there. The performance is definitely there. If the Raptor's
reliability proves comparable to the competition and if the
infrastructure/support hardware surface, WD will have a viable contender.

----------------------------

Ken Meyer
Not a Guru, but an Avid Researcher


-----Original Message-----

From: k12osn-bounces at redhat.com [mailto:k12osn-bounces at redhat.com]On
Behalf Of Jim Kronebusch
Sent: Monday, September 20, 2004 6:15 AM
To: 'Support list for opensource software in schools.'

Subject: RE: [K12OSN] SATA vs SCSI

> In an earlier thread, someone mentioned using SATA drives successfully
> with a large number of users, 25+

> When I go to configure next years server(we will upgrade) can I use SATA
> drives in a raid 0 configuration?  I want to increase performance hence
> the raid 0, but heard that SATA drives might not handle a class of 25+
> workstations

In my opinion I would always go RAID 5 whenever possible.  You get the
speed of striping and the redundancy of mirroring...but better.  As far
as SATA RAID goes over SCSI it is definitely cheaper.  But take my
advice and post before buying and controller and make sure you talk to
someone who has one running and booting before buying.  I didn’t do that
and now my nice RAID 5 SATA controller is only running /home and is
unable to boot, so my boot drive is a 20GB IDE as a quick fix.