[conspire] Last Year's Supercomputer

Rick Moen rick at linuxmafia.com
Mon Jun 9 19:26:55 PDT 2003


Quoting Edmund J. Biow (biow at bigfoot.com):

> I'm honored, Suh!

Credit where due.

> I try to buy dirt cheap, but necessarily bottom of the barrel.  Those 
> $10-$30 after-rebate burners you see advertised from big box stores are 
> generally sold by companies that rebadge drives from a variety of makers 
> (e.g. Verbatim, Cendyne, Pacific Digital, etc.)  Often you can deduce 
> from the serial numbers on the boxes which drives are relatively higher 
> quality (Lite-On, Acer) and which are to be shunned (BTC, Optorite).  

Take note, folks.  He's right.  (I can verify what he says about Acer,
and will take his word about Lite-On.)

The major brands (Ricoh, Philips, Yamaha, Sony, Toshiba...) have been
known, as well, to OEM their models behind obscure off-brands as well as
less obscure ones.  A bit of research _or_ just scrutiny of the unit
will often turn this up.  What you may never know is the background
story of why that OEM deal happened.  It might have been a production
batch with really bad quality control that the manufacturer doesn't want
to put its name on, so the batch was jobbed off to someone else.  Or
something else of that sort.

I dunno; I seem to do pretty well _with_ brand pickiness by just biding
my time and buying at the right time and place.

> On the economic front, remember that if you go SCSI you also have to 
> shell out for a SCSI card and sometimes for cables.

It helps to have been accumulating this stuff for ages.  Other people
don't know the value of components when they throw away old servers, for
example, which is how I got two Adaptec AHA-2940U host adapters sitting
out in the garage box o'parts waiting to be needed again.  Even better,
someone was throwing away a Symbios / LSI Logic SYM8951U card (53C895
chipset), which does 40 MB/sec _and_ supports LVD mode.  Too bad none of
those supports a wide data path; I have two that do, already deployed in 
knockaround machines.  The ones in the garage box are there because
they're not as good.  (I also have two ISA Adaptec AHA-1542B cards,
good enough for non-HD peripherals, but only really old machiens still
have ISA slots at all.)

Good internal SCSI cables?  I have dozens and dozens of quality ones of
(I think) both narrow and wide varieties, which $FIRM was throwing in
the dumpster one day for no good reason.  (Need a few?)  I also have a
couple of rolls of bulk cabling (cut to suit) and crimp-on connectors
for my old WAN/LAN consulting business, bought cheaply at Graybar back
when.  Why buy expensive pre-made cables, especially at Fry's prices?

So, on the whole, cost of both host adapters and cabling hasn't been a
factor for me.  And I should _particularly_ stress that my acquiring
most of those parts for free wasn't rare or a fluke:  Lots of people I
know do likewise, all the time, mostly benefitting from corporate sites
that couldn't find quality parts with a map and flashlight, who were
throwing them away.

> Also, according to Pricewatch the cheapest SCSI burner is a TEAC
> 12x10x32 for about $113 after tax and shipping & the prices climb very
> rapidly from there.     You can get an Acer 48X drive to your door for
> ~$38, no rebate, about a third of the price.  And you'll almost never
> find a mail-in rebate deal on a SCSI part.  So even if the IDE burner
> dies in a year or two, the way tech prices drop one could probably get
> a faster replacement drive for less than half of the money saved by
> not going SCSI & still have enough money left over for a "pretty good
> weekend in Vegas," in the immortal words of Slim Pickens in Dr.
> Strangelove.

Well, I'll bear that in mind if/when the Ricoh eventually dies.  But
it's worth a little for me just to run a system with much simpler
software configuration, with no ide-scsi shim driver, no total system
freeze-ups when bugs in that driver cause it to hang while in I/O mode, 
no funky arguments that need to be provided to the booting kernel, and
no gyrations needing to be worked out every time some new factor (such
as devfs) gets introduced.

You see, I have a working theory that the elimination of technological
bullshit has non-zero economic value that manifests primarily as greater
satisfaction and peace of mind, despite the fact that you cannot list it
on eBay.  ATAPI CDR/CDR-W and tape drives aren't near the top of my list
of such bullshit implementalia, but _are_ comfortably ensconced
somewhere in the middle.  

> [...] older Socket 7 and Slot 1 boards often only had 3 PCI slots and
> a similar number of ISA slots, and you needed at least one slot for
> video and possibly others for NIC and sound.

Ah, here you're on ground more intimately familiar to me than, say,
Athlon XP CPUs.  

One of the things most important to system planning has always been, of
course, making sure you have adequate maneouvering room for the medium
future.  For that reason, a Socket 7 or Slot 1 mainboard with only 3 PCI
slots -- unless it also had at least one separately usable AGP one or
had good-quality ethernet and/or sound built in -- was inherently
unacceptable merchandise for general-purpose computing.  For example,
even my cheapo (1997 vintage) Socket 7 FIC PA-2007 has four PCI slots.
(This was Baby AT form factor, 1 MB onboard level 2 cache, VIA Apollo
VP2 chipset aka AMD 640.)

However, back then, you could even get by -- barely -- with one of those
3-PCI board if it also had a few ISA ones, as does my PA-2007, because 
most of us back then had a few good ol' Creative Labs Soundblasters of
various vintages in the parts pile.  My garage box still has an AWE64,
an AWE32, and a couple of Soundblaster 16s -- all still perfectly fine
_if_ you still have ISA slots around.

> Sometimes even if you have an extra slot, mustering up an interrupt
> can be a problem.

Actually, no:  This is one of those problems that was nearly unique to
ATA ("IDE") fans of that era.  Why?  Because they kept having to add
special adapter cards (each grabbing an IRQ) for peripherals (scanners
zip drives, and tape drives, usually -- though ATAPI eventually fixed
some of that) the likes of which I would just hang off a SCSI chain.
I can honestly say that I've never come even close to running out of
IRQs on a SCSI-based system.

> These days IDE "Burn-Proof" or its kindred works quite well with 
> half-way decent media.  

What amazes me is how long people put up with the problems these finally
somewhat address, and how grateful they become about finally being able
to decently multitask.  Back around 1995, I was burning CDs with no
worries whatsoever on my 486DX2/66 all-SCSI Linux server, which was
simultaneously being a public Web/ftp server, ssh shell-account server
for several dozen people around the world, mailing list (majordomo) and
SMTP host, _and_ X11 server for my local desktop needs -- all with 64 MB
total RAM.  (Mind you, I used fvwm2, not Enlightenment.)

> My cousin was really surprised that we could burn 3 copies of Mandrake
> 9.1 under RH8 with X-CD-Roast while surfing the net and downloading
> the other ISOs with gFTP without a buffer underrun, even with a
> several year old Yamaha 8x burner without Burn-Proof. 

Welcome your cousin to the mid-90s for me.  ;->

> Likewise, these days 7,200 RPM IDE hard drives have performance that is 
> much closer to that of SCSI drives than it used to be (reliability is, 
> perhaps, another matter entirely), and the new S-ATA technology is very 
> promising.  

Yeah, they've been saying that about consecutive generations of ATA
since the early 1990s.  Here's how things were in 1996, for example:
http://linuxmafia.com/pub/hardware/eide.txt.gz

I readily believe that ATA technical standards keep advancing.  They
absolutely do.  This is a simple consequence of Moore's Law and general
improvements to electronics.  But ATA's always had a certain lack of
uniformity of standards observance and a tendency to boast terrific
numbers in areas that really don't matter (like theoretical maximum
bus transfer rates), while falling apart to some degree in multitasking
and multiuser real-world deployments.  

But that's the sort of thing you expect of something that exists to fill
the low-price niche:  ATA's always fundamentally been about hitting
bottom dollar.

> I saw a Seagate 80 gig serial ATA drive at Fry's for $120 
> yesterday vs. $160 for a 74 GB Maxtor Quantum 80 pin SCSI drive (again, 
> from Pricewatch).  

Please note that I _still_ wouldn't buy a Seagate drive on a bet, and
wouldn't even before they shortened their warranty period.

> Both drives have comparable data transfer....

Actually, no, they don't, really.  They have comparable data transfer
_figures_, which will mislead the unwary.  On modern drives, you cannot
saturate the bus -- of either type --  because data cannot be physically
read from the drive that quickly, even under contrived test conditions
of a completely quiescent single-user system reading ordered data from
consecutive sectors of a single cylinder.

Under more-complex usage patterns, intelligent disk activity becomes
increasingly a factor, such as the ability to reorder pending read tasks
into the order involving minimum seek activity.  These are the kinds of
areas where SCSI has traditionally shone.  Every year, I hear from some
ATA guy "Well, that used to be the case, but isn't any more", and so far
it's still been true.

> However for me HD reliability is much more important than for a 
> CD-RW drive.

That's why I like things like SCSI hotfix areas and related
hardware-level defect mapping.

> Good SCSI drives often come with 4-5 year warranties, while last year
> most makers went from 3 year to 1 year warranties with their IDE
> drives (you can still get 3 year warranties on some of the higher end
> models with the 8 MB buffer).  And SCSI has better seek times. 

You know, the difference _is_ just about wholly artificial, which is
maddening.  There's no reason the same drive mechanisms can't be offered
simultaneously at very slightly different price points depending on the
interface card you prefer.  The observed market differences seem to just
reflect a deliberate policy of market segmentation.  

I haven't had to buy drives in several years, since the days when 36 GB
was actually considered large.  It's a good thing, because the
artificial premium for SCSI -- which has increased in the last decade --
would really piss me off.  

That having been said, like most people who neither accumulate warez nor
MP3s nor movies nor pr0n nor run my own Debian package mirror nor
multiboot five OSes per box, I find that $100 IDE _and_ SCSI hard drives
both come in exactly the same size, one called "too big to fill and too
big to cheaply back up".  ;->  

> On the other paw, those 10,000-15,000 RPM SCSI drives can be amazingly 
> annoying, with their high-pitched whines.

This is one reason I like satellite computing via 802.11b-equipped
laptops.  The irony of that is that it means I use an all-ATA system.  ;->
But significant crunching and network tasks are on the whining all-SCSI 
VA Linux PIII-Katmai/500 running at a bearable distance in the living
room.  The laptop's mostly just an intelligent terminal.

But yeah, that noise problem's a bitch.  A lot of people don't realise.

-- 
Cheers,                    I've been suffering death by PowerPoint, recently.
Rick Moen                                                     -- Huw Davies
rick at linuxmafia.com  



More information about the conspire mailing list