COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	allian[w86,jmc]		Rationalization of buying Alliant computer
C00006 ENDMK
C⊗;
allian[w86,jmc]		Rationalization of buying Alliant computer

Please comment on the following draft which has not yet been sent
to Steve Squires.
squires@isi
computer purchase plan
Steve:

	We are tending in the direction of buying the computer for
QLISP from Alliant for the following reasons:

	1. It is M68000 compatible, and Dick Gabriel thinks that
Lucid can make Qlisp faster for it than for other machines.

	2. Jack Test, their systems person, knows Lisp well and is
strongly motivated to make whatever system modifications are required
so that Qlisp will run well.  I want to tie this down in any contract
we might sign with them.

	3. The recent reorganization of Encore left us with
doubts about their capablilities in the operating system area.
Mr. Pompa did not give the impression of being in charge technically.

	4. Going towards MIPS or some other RISC configuration seems
like a good idea, but this seems not to be immediately feasible.
It will, however, be just as easy to go to RISC from the 68020
architecture of Alliant as from the 32032 of Sequent and Encore.

	5. MIPS and the other available RISC machine (the IBM RT) both
have been designed as sequential architectures.  The Alliant has some
features oriented towards concurrency, and it will be important to
determine whether they are adequate and what improvements are needed.
Because the Alliant is a micro-coded machine, this may be more feasible
than with some of the others.

	It is relevant to make a few remarks on how I see the Qlisp
project in relation to longer term goals.  I view the first implementation
of Qlisp as a proof-of-concept and development tool.  The history
of computer science does not make it seem likely that we will get
a parallel Lisp implementation with a 20 year life on the first try.
Therefore, I put getting experience quickly at a higher priority than
moving quickly towards what will be a long term hardware-software
solution.  I hope that the Qlisp project can develop a system that
others will be encouraged to use.

	The hardware decision problem would be much less problematical
if there were more time to make decisions.  Stanford got a one year
extension of the DARPA hardware contract, and it's up in May.  Could
you look into the possibility of another extension at least for the
parallel machine decision.

John McCarthy