Re: tlug: Karl-Max has cool dreams [was: dual-pentium processors]

["Stephen J. Turnbull" <turnbull@example.com>]

>>>>> "Manu" == Manuel M T Chakravarty <chak@example.com> writes:

    Manu> I guess you are not really often above 1, 2, or
    Manu> maybe 3 processors.

    >> Never, unless I'm running a compilation and cthugha at the same
    >> time, but note the cost assumption I made: $/processor then =
    >> $/byte today.  Also, I'm running what is basically a glorified
    >> word processor.  I don't use multimedia (especially not
    >> audio/video input), do image processing, etc, etc.  But suppose
    >> it becomes cheap to put a processor on each of those tasks?

    Manu> I am sorry, but actually, your approach of usin many small
    Manu> and cheap processor was already tried out.  The company was
    Manu> Thinking Machine and the machine the CM-2.  The company had
    Manu> to face Chapter 5 a couple of years ago.

But you're still thinking about a single task.  (So is Karl-Max.)  I'm
not.

    Manu> A car is a distributed processing system, not a parallel
    Manu> system.

So what?  The issue is "symmetric multiprocessing" which puts some
limits on the form of the processor (processors must be
similar/identical) but not on the task.  So I'm choosing to look at a
relatively loosely-coupled set of tasks, rather than a single
well-defined one.

    >> Or page-at-a-glance OCR to give a practically important
    >> example.

    Manu> Actually not that easy to parallize, because when it comes
    Manu> to the high-level stages like feature recognition etc.

Sure it is.  Typical page is say 40 lines of 60 characters, once
you've broken that down you can put each of 2400 processors to work.
By today's standards that is massively parallel.  The task of breaking 
down the page is also parallelizable, to some extent anyway.  Of
course, the approach I'm talking about is obviously one of the
embracingly parallelizable bunch.  But it is practically interesting.

You could apply this to face recognition in CUCMe, too.

What the hell, let's think really small---assume that you could put a
Pentium in a flat-screen pixel.  What would happen to multimedia then?

-- 
University of Tsukuba                Tennodai 1-1-1 Tsukuba 305-8573 JAPAN
Institute of Policy and Planning Sciences        Tel/fax: +1 (298) 53-5091
--------------------------------------------------------------
Next Nomikai: 18 September, 19:30 Tengu TokyoEkiMae 03-3275-3691
Next Meeting: 10 October, Tokyo Station Yaesu central gate 12:30
--------------------------------------------------------------
Sponsor: PHT, makers of TurboLinux http://www.pht.co.jp