RE: Re: Two CPU AMD PC and Mathematica
- To: mathgroup at smc.vnet.net
- Subject: [mg29636] RE: [mg29607] Re: Two CPU AMD PC and Mathematica
- From: "Wolf, Hartmut" <Hartmut.Wolf at t-systems.de>
- Date: Fri, 29 Jun 2001 01:36:09 -0400 (EDT)
- Sender: owner-wri-mathgroup at wolfram.com
Hello Orestis,
hesitating to enter that discussion, I just like to add a little point:
there are certainly many occasions where parallel processing dispatched to a
bunch of processors might be possible and result in considerable speedup.
The decisive question is, can and how can the system decide on doing that
and fork the thread of execution. That question is not at all easy!
First you must not destroy the semantics of the language. And part of the
basic semantics in Mathematica is the evaluation sequence, and that is
prescribed exactly, stating e.g. that in an expression
h[e1, e2, ...]
to *first* evaluate the head, *then* each expression e<i> *in turn*, *then*
apply transformations associated with attributes,... See §2.5.4 pp
Each of the steps can influence further processing downstream the pipeline,
e.g. change/add/remove global definitions. This also tells that Mathematica
is not a pure functional programming language where, lacking any side
effects and any variables designating store, we could map in parallel.
Consider e.g. In[2]:= i = 0; Table[++i! , {5}]
So for parallel processing code has to be inspected carefully. Of course you
could use Mathematica for doing that, and it could be very interesting and
perhaps rewarding! We might be curious to learn what Roman Maeder did, but
damn, I missed the opportunity to participate beta testing his package.
-- Hartmut Wolf
> -----Original Message-----
> From: Orestis Vantzos [mailto:atelesforos at hotmail.com]
To: mathgroup at smc.vnet.net
> Sent: Thursday, June 28, 2001 11:28 AM
> To: mathgroup at smc.vnet.net
> Subject: [mg29636] [mg29607] Re: Two CPU AMD PC and Mathematica
>
>
> The single most common Mathematica operation I use is Map
> (and variations of
> it, such as Table)...I don't see why it couldn't be made to run
> concurrently. Divide the elements of the list to be operated
> on and spread
> among available processors, or something like that. I figure
> there would be
> an impressive speed up of my programs, if this could be done...
> Orestis
>
> "Jens-Peer Kuska" <kuska at informatik.uni-leipzig.de> wrote in message
> news:9hc8d2$40r$1 at smc.vnet.net...
> > Hi,
> >
> > yes -- but in real life it is hard to get a speed gain,
> > even with an MPI program ;-)
> >
> > Usual it is better to write a MathLink program for numerics
> > and pure symbolic algorithms that can be paralleized are very
> > rare.
> >
> > Regards
> > Jens
> >
> >
> > Orestis Vantzos wrote:
> > >
> > > I have to agree...has anyone tried Maeder's Parallel
> Processing package
> with
> > > the aforementioned setup though?
> > > Orestis
> >
>
>
>