Return-Path: <foner@media.mit.edu>
Received: from pacific-carrier-annex.mit.edu by po11.mit.edu (8.9.2/4.7) id XAA08378; Sun, 19 May 2002 23:26:03 -0400 (EDT)
Received: from out-of-band.media.mit.edu (out-of-band.media.mit.edu [18.85.16.22])
	by pacific-carrier-annex.mit.edu (8.9.2/8.9.2) with ESMTP id XAA15150;
	Sun, 19 May 2002 23:25:22 -0400 (EDT)
Received: (from foner@localhost)
	by out-of-band.media.mit.edu (8.9.3/8.9.3/MEDIA) id XAA07739;
	Sun, 19 May 2002 23:25:11 -0400 (EDT)
Date: Sun, 19 May 2002 23:25:11 -0400 (EDT)
Message-Id: <200205200325.XAA07739@out-of-band.media.mit.edu>
From: Lenny Foner <foner@media.mit.edu>
To: marc@MIT.EDU
CC: tytso@MIT.EDU, jmorzins@MIT.EDU, sipb@MIT.EDU
In-reply-to: <t53offcgjc0.fsf@horowitz-m1.mit.edu> (message from Marc Horowitz
	on 19 May 2002 13:04:31 -0400)
Subject: Which hard disk vendor do you trust?
CC: foner@media.mit.edu

    Date: 19 May 2002 13:04:31 -0400
    From: Marc Horowitz <marc@MIT.EDU>

    "Theodore Ts'o" <tytso@MIT.EDU> writes:

    >> Also, most people don't leave their computers on 24x7, or if they
    >> do, the disks aren't in use all the time.  I've never had problems
    >> with Maxtor disks that have always been spun up, although granted
    >> the disks heads are generally far more idle than what you describe.

    I don't know what your application is, but there are consumer reasons
    to want drives like this.  In particular, my TiVo DVR has a pair of
    IDE drives in it, and they are running 24/7, because it's always
    recording in order to allow the user to rewind.

And not just TiVo's!  Sun V100 and Netra 1U servers take IDE drives.
I'd think that the sort of market segment that buys 1U servers would
be screaming bloody murder if their drives failed in months; certainly
if Sun were the only vendor to do this, no one would buy from Sun.
I'd also be pretty dubious of claims that servers don't run 24/7 and
don't spend most of their time thrashing their disk heads.

As for TiVo's... while not -every- disk I've lost has been in my TiVo,
the majority of them have been (at least 3; maybe 4?).  Interestingly,
the A drive typically burns them out most; I can't remember if the B
drive has done so, even though both drives started out as identical
Maxtor 80GB drives from the same batch.  (This matters because the B
drive, if present, stores only video; A stores video, plus both sets
of Linux partitions, plus the programming data & metadata describing
where video is; this means that the A drive is -constantly- seeking
because the TiVo is constantly running background tasks that touch
that information.  The live-video buffer may also be there, which
means that if the TiVo isn't recording a particular program (which has
a 60% chance or so of being on B, assuming the TiVo is fair about
space allocation), it's also writing to A to do its half-hour
live-video buffer.  The A drive gets a hell of a workout.)

[TiVo doesn't ship with these drives; mine shipped with a 15GB Quantum.]

[See also long digression at the end of this message.]

    I've got the original factory Quantum drive, and an add-on Maxtor
    drive; both have been working fine for 9 months or so.

I'll bet your Maxtor is drive B, yes?  I'm willing to bet, given my
experience, that if you swapped them (tools exist that would allow
you to do so in a few minutes, assuming you didn't mind losing stored
video), your new A drive (Maxtor) would die quickly.  At least, if
you have the same sort of luck I've been having.

    You might consider if your failure is not due to the drive, but due to
    the environment.  If you're having multiple drives fail, I would
    suspect heat problems.  If you turn your equipment off right now and
    touch the hard drives, how hot are they?  Perhaps you should consider
    adding more fans or otherwise improving cooling for your system.

The very first time I took a TiVo drive failure, I instrumented the
repaired setup thoroughly, taping a digital thermometer probe (with
peak hold on the thermeter) to the top face of the disk enclosure.
And I left it there for weeks, with the case closed, just in case
there was some rare event causing the fan to stop spinning or
something like that.  The TiVo motherboard sensor has never claimed a
temp exceeding 43C (I checked the logs, and I've also never seen an
on-screen display with a higher temperature.)  The TiVo has a
variable-speed fan, and mine typically spins at the lowest speed, so
even if the house hit high temperatures (it's occasionally gotten to
30C at home---but not ever with the drive that just failed on me), the
fan would just spin more.  My memory is that drive temperatures never
went near the top of the spec (55C).  Since I can't find those notes,
I just did it again, and discovered that a motherboard temp of 43C
(right now) corresponds to a disk temperature of 39.3C---more than 15C
below the drive's rated maximum operating temperature.

The grossly uneven failure rate of the TiVo A and B drives, btw, also
seems to eliminate thermal considerations (they sit side-by-side, and
measurements reveal they run at approximately the same temperature),
and (perhaps?) also the suggestion from one person that maybe a
marginal power supply is causing failures---that's a hard one to
evaluate, but it's hard to imagine how -any- power supply weirdness
could lead to a drive so damaged that it goes click click click...
(It's only on a few sectors---perhaps just two in my latest failure---
but if it was simply a matter of the drive writing bogus data during
some "event", it would have spared out the sector and continued.
This acts like it -can't- spare it out.)

Btw, everything electronic in my house is on UPS's, so it's not
spikes, sags, or other things like that, either.

[Long digression.]

What's been most fascinating about this discussion, especially on
another list that I sent this to, has been the number of people who
simply assumed that they could ignore my stated requirements and
wander off into SCSI-land.  I deliberately didn't mention the TiVo
when I started, because I figured it wasn't relevant information
(besides, I've lost at least one disk in an ordinary PC recently),
but I've been amazed at the number of people who seem to assume
that -of course- it's always possible to go to SCSI if necessary,
and hence didn't even talk about which might be the most reliable
IDE drives---they just assumed that SCSI was the answer, and some
got positively unpleasant about being asked to stick to the original
question and not assume that the requirements I started were
irrelevant.  (Note that I'm not talking about any particular reply
here; the most unpleasant of the flamers have responded in private
mail, and not from this list at all.)

Obviously, in the case of products like a TiVo, you can't just go to
SCSI.  Also obviously, it cannot be the case that TiVos are simply
dying in 3-6 months, or TiVo wouldn't be in business any more.  They
currently ship with some rather large drives (60GB?), though typically
with smaller drives, e.g., 15GB; it's been proposed that smaller
drives are more reliable, though no one has actually pointed me to
real data about it, or why this might be so.

If the much-touted reliability of SCSI is not a myth (does anyone have
pointers to real research, hopefully w/associated URLs, in which
people have studied this under controlled conditions?), that's a very
interesting bit of market segmentation.  Because the controllers used
to cost more, SCSI drives were assumed to be less price-sensitive,
thus building disks that lasted longer was also presumed to be okay
even if it cost more.  This meant that IDE drives were allowed to go
to hell, quality-wise, leading to a positive-feedback loop where, even
though there's now no reason to prefer SCSI from a cost standpoint,
it's become accepted dogma that SCSI is more reliable---and thus users
are willing to buy SCSI controllers and pay extra for SCSI drives to
get that reliability.  But is there any truth to the mythos?  Even
though SCSI won't help me, I'd be interested in seeing rigorous,
controlled studies that might elucidate it.  I'd be -especially-
interested in -recent- data; while I heard about the whole cheap-IDE
stuff years ago, does anyone know if it's still true?  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?
