In his e-mail message of 1993/02/17, Frederick Ford wrote
I have an opportunity to get some "free" time on either a super
computer or a parallel computer.
The super computer options are:OS C compiler IBM ES-9000 AIX/370 AIX/370 C compiler IBM RS-6000 AIX AIX XL C compiler/6000 Cray-YMP UNICOS Cray standard C compiler
The easiest option is the RS/6000 (usually those machines are considered
workstations or servers, not super computers). Porting GAP to the
ES-9000 (under AIX) should also be reasonably simple. I would expect a
ES-9000 to run GAP about a factor of two faster than a large RS/6000.
Which machine is better depends on the number of other users on each
system, the amount of memory installed, etc.
GAP does *not* run on Crays. The reason is that GAP makes certain
assumptions about the format of pointers that aren't valid on Crays. It
could probably be made to work, but (as far as I know) nobody is
currently trying. It wouldn't make much sense because GAP could not make
use of the special hardware (vector units, etc.) that make Crays so fast
for numerical code. So GAP on a Cray probably wouldn't run significantly
faster than on a fast UNIX workstation (maybe twice or three times as
The parallel computer is a Connection Machine (massively parallel I'm
told). The OS is System V, BSD 4.3 compatible. I don't have any
details on the C compiler version, but it's whatever Thinking
Machines is distributing as their "standard" C compiler.
GAP contains no explicit parallel constructs, and I doubt that there are
many place where the compiler could automatically find parallelizable
code. I don't think you should try.
-- .- .-. - .. -. .-.. --- ...- . ... .- -. -. .. -.- .- Martin Sch"onert, Martin.Schoenert@Math.RWTH-Aachen.DE, +49 241 804551 Lehrstuhl D f"ur Mathematik, Templergraben 64, RWTH, D 51 Aachen, Germany