3.576 supercomputing the humanities, cont. (174)
Willard McCarty (MCCARTY@vm.epas.utoronto.ca)
Thu, 12 Oct 89 19:54:38 EDT
Humanist Discussion Group, Vol. 3, No. 576. Thursday, 12 Oct 1989.
(1) Date: Wed, 11 Oct 89 18:28:56 CDT (39 lines)
From: "Michael S. Hart" <HART@UIUCVME>
Subject: Re: 3.567 supercomputing the humanities, cont. (73)
(2) Date: Wed, 11 Oct 89 22:04:22 EDT (47 lines)
From: amsler@flash.bellcore.com (Robert A Amsler)
Subject: Supercomputers and numbers.
(3) Date: Thu, 12 Oct 89 13:53:00 EDT (38 lines)
From: Tom Thomson <tom@prg.oxford.ac.uk>
Subject: Re:3.567 supercomputing the humanities, cont. (73)
(4) Date: Thu, 12 Oct 89 12:27:00 EDT (20 lines)
From: "Vicky A. Walsh" <IMD7VAW@OAC.UCLA.EDU>
Subject: Re: 3.567 supercomputing the humanities, cont. (73)
(1) --------------------------------------------------------------------
Date: Wed, 11 Oct 89 18:28:56 CDT
From: "Michael S. Hart" <HART@UIUCVME>
Subject: Re: 3.567 supercomputing the humanities, cont. (73)
All computer functions are numeric. However, it *is* true that certain
chips and cards are made expressly for floating point and other types
of calculations. Nevertheless, the price of distributing, creating,
editing, and all the other functions of publication are *ALL* quite
a bit less expensive than the equivalent paper publications.
For the price of any library, an electronic library can be constructed,
complete with backups, terminals, and sufficient portals to release
materials on Mac, IBM, or other media.
A super-library would certainly justify the expense of a super-computer.
I, myself, run our library on a pair of 386s (one is a backup) and SCSI
hard drives which can hold 10G per controller, and multiple controllers
may be installed. While this is super-computer enough for my uses, and
can hold ALL the data collected by ALL the electronic text repositories
of which I am aware, it is NOT the kind of computer I would want to run
the entire library of any major institution.
Michael S. Hart
P.S. Some of you may not realize that the huge mainframe computer that
was used to create the first of the Project Gutenberg texts, had a core
memory of 64K, and a handful of removable 5M drives, and cost $millions
in installation and maintenance. Our current computers require no more
room, no air conditioning, and have between 640K and 8M core memory and
drives as mentioned above. The time interval of this change is not two
decades, and not even half a decade since the acquisition of our IBM-XT
which took us out of the mainframe world. Intel has already released a
spec sheet for the 80787(?) due to be released in the year 2000: which
will contain 4 processors with caches, and will need no math processor.
Enough for now, as someone recently quoted - new ideas do not succeed a
year or a decade earlier by convincing opponents, they succeed because-
and I hesitate here, but don't want to eliminate the punch line of that
quote-the opposition dies off.
(2) --------------------------------------------------------------61----
Date: Wed, 11 Oct 89 22:04:22 EDT
From: amsler@flash.bellcore.com (Robert A Amsler)
Subject: Supercomputers and numbers.
It is true that supercomputers (and in fact ALL computers) are
inherently numerical. However the numbers are 0 and 1 and not
0,1,2,3,4,5,6,7,8,9.
One of the things supercomputers are best at is converting numbers
into images. Scientific visualization is all the rage these days
as physical scientists discover that it is FAR easier to look
at a pretty full-color image rotating in 3-space than to
stare at 1000 pages of numbers. ``This is NOT your father's
supercomputer...''
Don't become trapped in the belief that `numbers' are an end point
of anything (except perhaps number theory.... :-)). Numbers are a
tool invented to manipulate models of the natural world. It is
the natural world that scientists are after, not the numbers themselves.
So... they are using the computers of today not JUST to calculate, but
to convert those calculations into imagery (which afterall, is
the best way to communicate to human brains evolved in a natural
world themselves).
It is true--lest someone complain I am distorting the answer, that
what people most often develop for many supercomputers are things
like floating-point accelerators to really zip those digits around;
and that the languages available in which to program many
supercomputers are more likey to be FORTRAN than SNOBOL4, but I would
STILL claim that every discipline can benefit from supercomputing if
they think carefully about their intractable drudgery problems and
come up with the right reformulations. Putting together fragments of
ancient writings isn't fundamentally different than assembling
genetic codes or breaking cyphers--the latter of which certainly use
supercomputing. Figuring out which hand-cut type characters were
used in what order in printing incunabula by digitizing the book's
characters and comparing features is the right stuff for
supercomputing. There are lots of tasks which `could' be made
supercomputable if one seeks out the right open-minded supercomputing
scientist and asks them to think about your humanistic computing
problems. However, just try to remember that supercomputers are only
the CPU--not the peripherals nor the network. If what you really
want is not computation, but the associated operations before or
after computation, then a supercomputer can't really speed things
up that much. Jet planes can't really get you to the airport or
check your luggage any faster, even if they are the Concorde.
(3) --------------------------------------------------------------40----
Date: Thu, 12 Oct 89 13:53:00 EDT
From: Tom Thomson <tom@prg.oxford.ac.uk>
Subject: Re:3.567 supercomputing the humanities, cont. (73)
Willard's note (supercomputing unrest) of 11 Oct makes two points which I think
are quite wrong.
First, supercomputers are inaccessible and unfriendly, just as mainframes used
to be: in fact things have moved on quite a bit since the early sixties when
mainframes were fed their diet of perforated paper (tape or card) by
white-coated acolytes and only initiates were allowed to see the beast; we have
a better understanding of how to build user interfaces, a better understanding
of how to split work between central and distributed components of a computer
system, and far better networking capability which allows us to provide high
bandwidth communication on the end-user's desk. The mainframes (or maybe they
are "supercomputers") which I use are no more faceless, distant, or unfriendly
than the PC on my desk. If your local "supercomputer" has this problem, moan at
the people who run it/provide it and get the problem fixed, don't give up
without a try.
Second, supercomputers are designed to be super-fast for numerical work: well,
just try doing numerical work with something like an ICL CAFS engine - and then
try doing some complicated searches for strange word patterns in the
Shakespeare first folio! You'll find the former not just difficult but
completely impossible while the latter is quite straightforward.
Most research into supercomputers today is oriented towards symbolic
computing rather than numeric.
If "humanists" bow out because they feel that supercomputers are there just for
those who want to discover yet bigger prime numbers or do finite-element
integrations over finer meshes than ever before, no-one will develop the
software needed to apply supercomputers to the humanities: this will end up
being a fine example of a self-fulfilling prophecy. Fight for the facilities
you want or at least for the tools that will allow you to develop the
facilities you want without spending any significant time on software,
rather than letting the physicists have all the computing resources!
Tom Thomson tom@prg.ox.ac.uk, tom@stl.stc.co.uk
(4) --------------------------------------------------------------108---
Date: Thu, 12 Oct 89 12:27:00 EDT
From: "Vicky A. Walsh" <IMD7VAW@OAC.UCLA.EDU>
Subject: Re: 3.567 supercomputing the humanities, cont. (73)
I would like to reply to the last part of Willard's comments on supercomputers.
Yes it is true that some supercomputers (mainly the Japanese ones, I believe)
are more single purpose than other computers we use. However, all mainframes
were designed to do number crunching, and supercomputers more so, BUT
all data are bits to the computer. The advance of the supercomputer is not
only great speed and large memory, but vector processing and parallel
processing both of which can be used quite nicely for many humanities
applications. It works so well for mathematical stuff because it does
many many iterations of the same thing at the same time. But many
language applications could benefit from this same concept.
I know from personal experience that Cray Inc. was very reluctant to admit
that other things could be done on its machines but as people demanded
applicatons they became available. They have already been asked about
humanities applications, and if more of us ask they will come.
Vicky Walsh, UCLA