Fwd: parallel computation was: Re: question



	

	

	

	

On 7/22/2011 8:53 AM, Paul Cefola wrote:

>  Is anyone thinking about the possibilities for applying maxima
>  symbolic algebra system in current computer architectures with
>  multi-core CPU and multiple processing element GPU?
>

type  parallel algebraic computation  into Google.
The first page of hits shows you
  a list of papers on libraries, and collections
of papers in conference proceedings on this topic.



Parallel computer architectures have been around for several decades.
See
http://pasco2010.imag.fr/

They were not so cheap back in 1988, but the good ideas and not-so-good
ideas could be explored, and people wrote about symbolic computation,
tree-search stuff, interpolation-evaluation stuff, parallel/distributed
hash tables, back then.

I think the best payoff is to translate a symbolic problem into
one of the classic numeric tasks, and use the best available systems
for those.  That way when someone invents a better FFT, your own program
runs faster too.

There are also language extensions in various lisps supporting parallel
computation,
but if Maxima is to run on any of a number of lisps, one has to write in
the common subset, making it harder.  So far as I know, this has not been
attempted.

You can read a lot about past experience, and some of it is, I think,
relevant.  See the PASCO proceedings as a start.

To the extent that symbolic computation uses an unpredictable but
probably large
amount of memory in some linked-list irregular structure, parallel computing
is tough.

RJF