I am running some pretty heavy Common Lisp, which currently takes over 1
week to run on a Mac Quadra 950 (fast 68040) using MCL 2.0. Can anyone
please suggest a hardware solution (cost is accepted!) which will run at
10-20 times the speed?
It has been suggested that a TI Explorer might help - could anyone
suggest a good Explorer configuration (I currently work in 6 megs of RAM
for MCL, but it still does plenty of GCing) and whether this is likely to
give the sort of speed increase that I desire, please?
Howard Oakley, EHN & DIJ Oakley, * ······@quercus.demon.co.uk
EHN & DIJ Oakley, Brooklands Lodge, * AppleLink UK0392
Park View Close, Wroxall, Ventnor, * CompuServe 70734,120
Isle of Wight UK PO38 3EQ * Tel +44 983 853605, fax 853253
From: Rob MacLachlan
Subject: Re: Speed freak seeks help
Date:
Message-ID: <C0179C.oz.1@cs.cmu.edu>
In article <··········@demon.co.uk> Howard Oakley <······@quercus.demon.co.uk> writes:
>I am running some pretty heavy Common Lisp, which currently takes over 1
>week to run on a Mac Quadra 950 (fast 68040) using MCL 2.0. Can anyone
>please suggest a hardware solution (cost is accepted!) which will run at
>10-20 times the speed?
One thing you should do before getting more hardware is to figure out
what is limiting the speed of the program. If, for example, your
program is floating-point intensive, then you could see a subtantial
gain from using a Lisp which open-codes float operations. Or perhaps
you are garbage-collection limited, in which case you want a
generational GC.
You also need to think about how memory intensive your application is.
High speed microprocessors are very sensitive to cache performance, so
if your working data doesn't fit in the cache, you won't see the sort
of speedup you would expect from the clock speed increase.
>It has been suggested that a TI Explorer might help - could anyone
>suggest a good Explorer configuration (I currently work in 6 megs of RAM
>for MCL, but it still does plenty of GCing) and whether this is likely to
>give the sort of speed increase that I desire, please?
I'm not up do date on the Explorer product line, but my hunch is that
you will see maximum speed from a top-end Unix workstation running a
good commercial CL implementation. Workstation performance increases
so fast that any special purpose Lisp machine is likely to be one
generation behind. The main advantage of the Lisp machines is they
have an excellent development environment and do run-time safety
checks at minimal extra cost.
You need *much* more memory to run a RISC Unix CL. I recommend at
least 32 meg.
Rob MacLachlan