Sender: marc@horowitz-m1.mit.edu
From: Marc Horowitz <marc@MIT.EDU>
To: nathanw@MIT.EDU (Nathan J. Williams)
Cc: Lenny Foner <foner@media.mit.edu>, tytso@MIT.EDU, jmorzins@MIT.EDU,
        sipb@MIT.EDU
Subject: Re: Which hard disk vendor do you trust?
References: <200205200325.XAA07739@out-of-band.media.mit.edu>
	<mtu4rh332wn.fsf@department-of-alchemy.mit.edu>
Date: 20 May 2002 00:35:03 -0400
In-Reply-To: nathanw@MIT.EDU's message of "19 May 2002 23:37:44 -0400"
Message-ID: <t53y9efh1xk.fsf@horowitz-m1.mit.edu>
Lines: 43
User-Agent: Gnus/5.0807 (Gnus v5.8.7) Emacs/20.7
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii

nathanw@MIT.EDU (Nathan J. Williams) writes:

>> > Why would vendors bother to implement two different lines---one more
>> > reliable than the other---for the -same- physical drives, then slap
>> > IDE interfaces on the less-reliable line?  Why not a single line
>> > that then bifurcates into which controller gets put on the drive?
>> 
>> Because manufacturing for quality is often an issue of tolerances, and
>> closer tolerances mean higher costs, in tools, maintenance, and QA. So
>> on the manufacturing side it makes perfect sense for there to be a
>> quality/cost trade-off. 
>> 
>> As has been observed in this thread, much of the market for IDE disks
>> is vastly more price-sensitive than quality-sensitive.  Therefore, it
>> also makes sense to slap SCSI controllers on the "close-tolerance",
>> more expensive disks, and IDE controllers on the "loose-tolerance"
>> disks.

I'm reminded of what happens if you measure the accuracy of resistors.
As you may know, you can buy resistors which are rated to be within
5%, 10%, or 20% of their advertised resistance.  (You can buy other
ratings, too, but those are usually different processes than the
5/10/20 process).  Of course, the more accurate resistors cost more.
One might guess that if I bought a number of similar resistors and
measured them all, they would fall in a gaussian distribution.  It
turns out this is not true.  If you buy a pile of 10% resistors, you
will find a gaussian curve which has been truncated outside 10%, which
is to be expected from the spec.  But you will also find that the
curve is truncated inside 5%.  The reason is simple: manufacturers
will make many resistors with a target resistance, measure the actual
resistance, and label them depending on how accurate they really are.
So, the less accurate resistors are, in a way, guaranted to be
inexact, because if they were more exact, they would be sold as
higher-quality parts.

Perhaps (and now I'm speculating) hard drive manufacturers have a way
of measuring how well-made a given drive is.  The "better" drives get
SCSI controllers and are sold for more, and the lesser drives get IDE
controllers and are sold at commodity prices.  This would account for
the claimed higher reliability of SCSI drives, while still allowing
the manufacturer to have a single production line.

                Marc