Re: IDE vs SCSI for RAID

["Stephen J. Turnbull" <turnbull@example.com>]

>>>>> "A" == A Sajjad Zaidi <sajjad@example.com> writes:

    A> But you have to admit that taking price into account really
    A> changes things. The fastest IDE may be slower than the fastest
    A> SCSI, but its a lot cheaper.

Sure.  There's always a price performance tradeoff.  Jc was implying
there wasn't; that IDE was everything SCSI is (for practical purposes)
and cheaper.  It simply is not.

    A> IDE would be a problem if you decided to use a master and slave
    A> device on the same controller, but it works great if you put
    A> one device per controller.

Depends on what you're doing.  Again, I don't know about modern UDMA
controller/drive combos, and it's been a while since I did a
head-to-head comparison (about 5 years)---but at that time a 486DX50 +
EISA + SCSI + 16MB running Linux 1.2.13, X, NCSA httpd, smail, and
Emacs gave acceptable interactive response even when I was getting
175,000 hits per day on the webserver.  A P120 + PCI + IDE + 32MB
combo didn't give better performance; in fact it was only acceptable
when I hauled the 486 out of retirement and stripped X and Emacs off
it, and let it run the httpd and smail (even though by the time I
switched to the P120 the web traffic was like 1/500th and the mail
traffic had halved).  I think the conclusion to draw is obvious.

    A> I had a file server with SCSI Quantums (SCSI) which started to
    A> die one by one (so much for reliability). so they were ditched

Sure.  The mechanicals are the same for the IDE and SCSI version of a
given drive, after all.  It's unlikely that they'd have significant
reliability differences _of that kind_.  But my first SCSI drive (a
full-height 4kg monster from Fujitsu) lasted 10 years of 360x24x7.
(Average of 5 days of power or network out at Tsukuba-Dai per year.
Life in the Third World, you know. ;-þ)  On the other hand, until the
_drive_ actually entered its death rattle, my /var/log/scsi was empty
except for startup messages.  I see disk farts and controller resets
from my IDE controller daily.

    A> them and replaced by a raid array (RAID-5) of 4 x 45GB IBMs
    A> (ATA100).

    A> The result was about 80GB of space and very fast access times,
    A> plus its been up for months without trouble.

Yeah, well, my ex-wife works for Quantum, that would explain _any_
reliability problems.  ;-)

But you're comparing apples and oranges, here.  It's ambiguous the way
you put it, but I gather your SCSIs were not RAID'ed.  Of course an
older set of ordinary SCSI drives are going to compare badly with RAID
array built from new drives.  Nor do you specify what variety of SCSI,
etc.  And how about the file systems?  There's a good chance that you
switched from ext2 to Reiser or something like that (you don't say so
I assume not, but that would make a difference too).

Also, how are you measuring the access times?  What really matters is
does throughput hold up when you're thrashing?  This is where IDE
tends to fall down.  If all that space is serving a couple score
humans running MS Office, the file server is not going to thrash.  If
it's trying to support an airline reservation system, you're gonna
have problems.

My point is not that you can't do a good job at a lower price with
IDE.  It's that it depends on how you use the system.  Me, I would
gold-plate the bus, the RAM, and the disk controller/drive combo.  CPU
speed etc is not as important for most applications, in particular not
for servers which are normally I/O bound.  In I/O bound applications,
SCSI definitely has an edge.

I have never regretted spending "too much" on my I/O.  I've always
found I needed more than I planned.  More bus, more controller, more
net, more modem, more memory to stay out of swap.  It's CPU speed and
similar stats that I don't get enough bang/buck from.  (I'm obviously
a candidate for a return to "big iron". :-)

More than the IDE improvements, though, USB will change the equation
since hanging things other than fixed disks (scanners, etc) off the
SCSI bus no longer makes as much sense.


-- 
University of Tsukuba                Tennodai 1-1-1 Tsukuba 305-8573 JAPAN
Institute of Policy and Planning Sciences       Tel/fax: +81 (298) 53-5091
_________________  _________________  _________________  _________________
What are those straight lines for?  "XEmacs rules."