FGCS Project Evaluation Report
Keith L. Clark
Imperial College
London, UK
5th June 1992
result and impact of the FGCS project will be its software, and its radically different
approach to developing parallel applications.
I believe that this will indeed be the lasting legacy of the project, rather than the features of
the PIM machines that have been built. (Did you really need to build five? Or was their
construction relatively inexpensive compared to the cost of building PIMOS, KL1 and its
extensions, and the applications?) However, computer architecture is not my field. A
comparative evaluation against stock hardware might prove the need for the specialized
architectural support of KL1. I understand that you intend to do such a comparison as well
as a thorough evaluation of the alternative Pims. A good result would be the identification
of a few low cost features that significantly boost the performance of KL1 and PIMOS on a
multiprocessor machine, as was mentioned by Takashi Chikayama in the workshop. Such
features could then be incorporated onto a general purpose commercial machine which
offered both UNIX and PIMOS. (Could PIMOS run as a subsystem of UNIX?) UNIX
would ensure initial penetration of the market and the ICOT software should then ensure a
runaway success for some ICOT licensed Japanese manufacturer.
Conclusions
The FGCS project is something of which Japan can be truly proud. It has
had more impact than any other research project in computer science. It was magnificent
and bold in conception, and has delivered much more than I expected it could achieve.
PIMOS, the Pim machines, KL1 and its extensions and the impressive
range of initial applications are significant achievements that are testimony to
the skill, dedication and single mindedness with which the goals of the project were
pursued. I suspect that in most other countries such a project would have ended long
before the 10 year deadline, either through withdrawal of government support or lack of
stamina of the principal investigators. ICOT and its associated researchers have also done
excellent research in other areas of LP, as evidenced by the many publications and the high
quality ICOT research report series.
The decision to freely distribute ICOT software is excellent, but this
distribution needs to be supported by good documentation of both the software and the
methodologies of its use.
You should definitely port KL1 and PIMOS to existing commercial
multiprocessor machines. In your achievements report you say "...the technology of
PIMOS as well as the KL1 language is .. applicable to most MIMD .. machines.. ". I
agree. By proving this you will increase the impact of the project. It is also necessary if
the freely distributed software is to be widely used for developing applications for parallel
machines.
You still need to convince a skeptical outside world that KL1 provides "..much higher
productivity and parallel program maintainability than any conventional language".
Document, perhaps also refine, your application development
methodologies. Describe the program level structure of your applications.
Develop more applications. Develop what Ehud Shapiro in his workshop
presentation called a `killer' application and which I referred to as a `demonstrator'. At the
outset of the FGCS project there was much talk of knowledge information processing as
the key application area of FG computers. Why not build a huge information processing
application to support management decision making? Such an application must be multi-
user, perhaps using KL1 and PIMOS implemented on a distributed loosely coupled
- 51 -