[**WARNING** This is starting to drift off-topic from Linux-Audio ]
> At work I've gone away from SCSI for three reasons - first is cost,
Yes, in cost per GB, nothing beats IDE. But for most sizes of drives, the
extra cost of SCSI isn't that bad. I've paid $90 for the last few 36GB
drives, which is quite reasonable really considering it's a 10,000 RPM drive
with a large cache and a good warranty.
> second is that the speed difference isn't very much anymore,
One thing I seem to notice is that many SCSI drives don't seem to use the
full platter in order to keep the heads over the "dense" zones of the media.
When I examine the transfer rate curves of my Fujitsu MAS 10K RPM drives,
for example, the difference between inner and outer edge throughput isn't
anywhere near as great as the difference in performance between the
inner+outer edges of my WD 200GB drive, for example.
For another thing... I know I can't prove this scientifically, but I have
ALOT of samples from different manufacturers and time periods over the past
15 years. My SCSI drives last a very long time, and when they've finally
kicked the bucket (rarely after less than 6 years), I've had ample warning
[LOG SENSE notifications of soft-retry counts, etc].
In contrast, my IDE drives have failed abruptly and without warning, many
times before the warranty was expired. I have a bucket of such drives,
which I intend to take the platters out to make attractive and very smooth
The IDE bucket has drives from Seagate, Quantum (the most common being the
Fireballs, an apt name if I ever heard one), and Maxtor. Oddly, I have yet
to throw a dead Western Digital drive into the Drive Bucket of Doom - but
that could be because I had a bias against the brand for a long time. But I
do have many running WDs in current systems as they tend to be the fastest
drives available for IDE. I remain offended that they are the only IDE
drive manufacturer which doesn't include a thermal sensor in their drives.
I have functioning SCSI drives in several machines from Seagate, Quantum
[dating back to the 1.6" 400MB PD-series, circa 1992 as well as a
hot-running Grand Prix Wide-Differential 4GB drive circa 1996], IBM, and DEC
[pre-Quantum acquisition]. This latter drive is a VERY BUSY Squid Proxy
Cache drive in my home network server [same one which does the audio logging
from the MERLIN phone system].
The last SCSI disk failure I had was a DEC 1GB drive (ca. 1993-94
manufacture), which began getting very slow and kicking up LOG SENSE alerts
[I use a hard disk driver I wrote myself for the Macintosh in this
instance]. It continued to function well enough to make a complete backup
of all of the files, and for another 2 weeks before finally expiring. The
drive was 6-7 years old at the time of death.
My bucket of dead ATA CD-ROM drives got so large, I surrendered it to a
landfill. As contrast, I've only retired two SCSI CD-ROM drives, and one
was a cheap Ricoh CD-R burner which I never trusted when it was brand new.
As an old SCSI hacker + wardog, yes, I'll admit some bias against IDE. But
I've given IDE alot of runtime as well as my SCSI stuff, and IDE continually
disappoints me, while SCSI just keeps on working.
I know the bare drives are mostly the same and the differences are in the
controller on the underside. I note that most SCSI drive warranties (3-5
years) are longer than IDE (which is now 1 year for most drives/models).
Perhaps the extra cost of SCSI (aside from the smaller capacity production
lines for the models) is due to better testing/more conservative designs, or
what have you. I wish I could really know for certain.
> third is that you can't get SCSI in Djibouti (or Honolulu for
Really? That's a bit surprising.
> I work with ship and airborne data collection/processing systems.
Those use hard disks? Fascinating. I'd have thought the environments were
too harsh. Your work sounds interesting, though, Jan.
> We've had problems getting
Ouch. That's amazing.
> Now we're starting to switch to SATA for the same reason.
UGH, SATA, at least in this first generation, is a mess. I've been
following some of the T13 standards discussions about it, and things don't
look very pretty there. All I can suggest is that you validate specific
drives to controller chipsets for the time being, and watch your cabling
lengths despite marketing b.s. which claims it can be "very long". Do not
violate the "grounding rules", and don't run SATA cables outside of
reasonably shielded environments if they have a great deal of radio emissions.
There are some publicly-accessible discussions of these issues, so this
isn't just FUD. See -- the author, Hale
Landis, is an involved member of the NCITS T13 [formerly X3T13] committee,
and has contributed a significant body of work to the ATA standards over the
A focus on Quality.