a good volunteer
Tue, 29 Sep 1998 12:15:29 -0400 (EDT)
On 29 Sep 1998, David Tillman wrote:
> Mostly dead. I was originally pursuing a hardware platform
> for the OS based on something not quite so brain-damaged as
> Intel. Hopefully a multiprocessor Motorola based system, but
> more likely something based on either the StrongArm or IDT
> The reality is that current PC platforms are too cheap to make
> a custom solution worthwhile unless some real performance wins
> can be demonstrated.
I think that's why Symbolics keeps going bankrupt. (Not that Open
Genera isn't worthwhile -- I haven't tried it.)
> (Hey Kragen, how about a Beowulf-in-a-box running LispOS?)
That's an interesting idea. Someone suggested turning the
Beowulf-in-a-box into a ccNUMA system today. Then you could just run a
multithreaded program and have it run across thirty CPUs. :)
Ideally, what you'd want for a multiprocessor box would be some
automatic, easy method of finding things that were possible to run in
parallel and would benefit from running in parallel.
There are some folks working on a 1-square-mm 1BIPS stack-based
microprocessor called the F21; their 500MIPS version should be back from
the fab in November. It has a built-in video card, sound card, network
card, parallel port, and a real-time clock. Stack-based CPUs are
probably better suited to running Lisp than register-based CPUs.
Main trouble is that they've done three fab runs so far, and none of
the results were working. The most recent one (from a year ago)
couldn't write to DRAM and couldn't do subroutine calls and returns.
The development company (www.itvcorp.com) has been able to do prototype
chip runs each month, and they've got chips now that they can actually
run things on. Their chips are called the 'i21', and they are closely
related to the F21.
(btw, the chip is being designed by the inventor of Forth.)
Information on its status is at <URL:http://www.dnai.com/~jfox/f21stat.html>.
<firstname.lastname@example.org> Kragen Sitaker <http://www.pobox.com/~kragen/>
A well designed system must take people into account. . . . It's hard to
build a system that provides strong authentication on top of systems that
can be penetrated by knowing someone's mother's maiden name. -- Schneier