[comp.arch] Non-Discoid Very Large Mass Storage Systems

mmm@cup.portal.com (Mark Robert Thorson) (05/03/90)

I'd like to know what technologies are competing with magnetic and optical
disk drives at the high end of the mass storage spectrum (high in terms
of capacity rather than speed).

In the old days (say, pre-1975), IBM had some weird thing that used canisters
of tape in a honeycomb array.  I think CDC had something similar.

Before that, there was the IBM 2311 data cell drive, Alan Shugart's first
really famous design.  A "data cell" was a magazine of tape strips, about
3 feet long and three inches wide.  The magazines were in a circular array.
A "picker arm" would reach into a magazine, grab a strip, and wrap it around
a drum.  After that, it would be read like a drum memory.  There was also
a mechanism to put the strip back in the magazine, I presume.
The machine's controller was a 16-bit minicomputer with 32K of core,
which was a pretty impressive computer (i.e. pdp-11 class) in those
days.

And there was the IBM 1360 Photodigital Storage System, a.k.a "Chipstore".
If I'm not mistaken, Los Alamos designed it, and IBM built three.  It used
tiny film cards, about 2.? x 3.? inches, which held 16384 somethings.  It had
gobs and gobs of these "chips".  In the late 70's, Lawrence Berkeley Lab
informed their user community that IBM was ceasing support for the machine,
and in the opinion of IBM engineers, LBL wouldn't be able to service it
themselves.  (The machine dates back to the mid-60's.)  It was such high
precision that a film chip written on one machine couldn't be read on another
machine.  LBL said they would shut down the writer one year before the
reader, in order to encourage users to move to other storage media (this was
announced in a message entitled "Enough Rope").  But when the time came to
shut down the reader, they decided to keep it available until it broke
down.  I never heard what happened after that.

What systems are comparable to these today, if any?  Does anyone know where
I can buy one of the antique machines, or significant parts of it, cheaply?
How much do you want for it?

dick@cca.ucsf.edu (Dick Karpinski) (05/10/90)

In article <29543@cup.portal.com> mmm@cup.portal.com (Mark Robert Thorson) writes:
>I'd like to know what technologies are competing with magnetic and optical
>disk drives at the high end of the mass storage spectrum (high in terms
>of capacity rather than speed).

I've been looking at various juke boxes and the new "digital paper"
devices.  Near as I can tell, whether rewritable or WORM, juke boxes
of optical disks run about $1/megabyte and cost about $35k to get
started in the game.  That cost/meg can be reduced to about 40 cents
if you go for humongous capacity and pay the $300k entry cost.  By
way of contrast, hard disks can be had for $3/meg and do not suffer
the 10 second or so access time.

Exabyte 2GB tapes can be had in juke box configuration with about
the same entry cost as the low end optical disk jukes giving the
cost/meg of the high end ones.

The new digital paper stuff seems to arrive first (about now) in
terabyte capacity and $200k+ pricing for mainframe channels.  One
open reel of 35 mm paper (800 meters or so) gives 28 second average
access time and 60 second end to end time.  Even though the medium
only costs $0.01 per megabyte, that comes to $10k for the reel.

More interesting to me is the 3480 cartridge tape lookalike system
expected to be introduced in about a year.  It is expected to lower
the entry cost by an order of magnitude and handle a mere 50 GB.
An inexpensive 10 cartridge changer brings you back to 0.5 TB.

The other nice entry level product is expected maybe in 1992 as a
fancy flippy-floppy with about 500 meg/side.  It'll have to compete
with ever cheaper rewritable optical drives and platters.  Here the
expected media cost of $10-20 beats current optical platters, but
then products you can't buy are often cheaper.

When last I tried to show the relevant statistics, I discovered 
that I am vitally interested in access time, transfer rate, capacity,
cubic feet per gig, footprint per gig, entry cost, media cost, 
reliability, etc etc.  Many too many things to show at once.

Dick
-- 
Dick Karpinski  Manager of Unix Services, UCSF Computer Center
UUCP:  ...!ucbvax!ucsfcgl!cca.ucsf!dick        (415) 476-4529 (11-7)
BITNET:  dick@ucsfcca or dick@ucsfvm           Compuserve: 70215,1277  
USPS:  U-76 UCSF, San Francisco, CA 94143-0704   Telemail: RKarpinski