Path: utzoo!mnetor!uunet!husc6!rutgers!ames!hc!beta!dzzr
From: dzzr@beta.UUCP (Douglas J Roberts)
Newsgroups: comp.lang.lisp
Subject: Re: Numerical Programming in Common Lis
Message-ID: <13334@beta.UUCP>
Date: 14 Dec 87 16:49:24 GMT
References: <477@PT.CS.CMU.EDU> <168700008@uiucdcsb>
Organization: Los Alamos Natl Lab, Los Alamos, N.M.
Lines: 28
Summary: We have found different results...

In article <168700008@uiucdcsb>, mccaugh@uiucdcsb.cs.uiuc.edu writes:
> 
>  I was intrigued by your remark that your "number-crunching" runs 100 times
>  slower on a SUN than on a Symbolics, and wa just curious to know if
>  1) your SUN lacked a numeric co-processor assist;
>  2) the Symbolics benefited from similar assists.
> 
We have also run comparisons between Symbolics and Suns. We have a
large discrete event simulation running on top of KEE (Knowledge
Engineering Environment) that was developed on Symbolics hardware (an old,
upgraded 3600 with an IFU). There is a fair amount of floating-point
computation as well as a lot of symbolic computation in this
application. When we loaded up the application on a Sun 3/260 we found
that it ran ~1.2 times slower (80% as fast) as it did on the Symbolics. 

I was therefore surprised that someone else had noticed a 100-fold speed
difference in LISPs between the two machines. We found that the Sun
run-time was very sensitive to the paging space and the application
memory image size. If either of these was too small, the application
would garbage-collect itself into the ground. Perhaps this was a
contributing factor to the above-mentioned 100-fold speed difference.

--Doug
-- 
---------------------------------------------------------------
			Doug Roberts
			dzzr@lanl.gov
---------------------------------------------------------------