Evaluation of the FGCS Project
David H. D. Warren
Department of Computer Science
University of Bristol
research and development achievements of ICOT are on a par with the three institu-
tions, MCC, ECRC and SICS, which are most comparable with ICOT and which are
representative of the very highest level of computing research in the world. Moreover it
should not be forgotten that those three institutions came into being largely following
in the footsteps of ICOT and the FGCS project.
Recommended Future Steps
I strongly recommend that ICOT's work should be continued in some form beyond
the 1993 official end date of the FGCS project. The nucleus of highly gifted people
and expertise built up at ICOT should not be allowed to evaporate, but should be
continued within a smaller and more flexible framework. The KL1 software should be
made available on widely available standard hardware, including Unix uniprocessors
and multiprocessors such as Sequent Symmetry and perhaps BBN Butterfly. The PIM
hardware should be examined to see whether it might potentially form the basis for
commercial products if standard languages and operating system were supported. More
effort should be put into evaluating the FGCS results, and especially in comparing the
performance and usability with the best conventional alternatives. Speedups and good
load balancing are not enough by themselves; one needs to show that applications
perform better than they would by other approaches with comparable implementation
effort. There should also be continuing research, especially in the areas of knowledge
processing and applications. I would suggest that all this would best be done within
a much smaller research institute, with selected long-term staff, and a focussed but
flexible ongoing research programme (c.f. for example SICS).
It is understood that MITI is anxious to have official overseas collaboration in any
extension of the FGCS work. My own group would be interested in collaborating
with ICOT (or its successor) in evaluating ICOT's parallel applications developed in
KL1, to see to what extent the same problems can be solved through more directly
declarative logic programs, and whether comparable performance and parallelism can
be obtained from logic programming implementations supporting implicit parallelism
(such as Andorra-I). Unfortunately, DTI (the UK counterpart of MITI) requires 50%
funding from UK industry for any research it supports. So long as the ICOT work is
only available on custom hardware, it is unlikely that UK industry would be interested.
And even if the ICOT software were ported to standard hardware, the likely payoff
from such research is too long-term for most UK industry (with its rather short-term
horizons). Therefore, I am afraid the chances of official UK involvement, through DTI,
in continuations of the FGCS work seem poor, for the near term at least.
- 148 -