Re: Multithreaded scientific programming

[Pascal Obry <pascal@obry.net>]

Francis Burton a �crit :
> In article <eg3c27$7kl$1@reader1.panix.com>,
> David B. Chorlian  <davidc@panix.com> wrote:
>> Please, let's have something worthwhile rather the above joke.
> 
> You don't say why it is a joke and not worthwhile. Is this
> an opinion, or fact?

Probably not a joke as this was presented by Jim as a "modest example".
I also think that this is not a proper solution as creating 9 tasks for
each 3x3 matrix product is a waste of resources. But isn't this the path
followed by OpenMP ?

First of all : I'm no expert on this domain.

I tend to think that in a real world scientific applications there is a
lot of vector/matrix/whatever computations to be done. Also not the
whole application is computation intensive (data loading or
initialization for example).

I like to think as a work flow. Some tasks are preparing data to be
computed and storing them in a queue. Some others are getting the job to
be done from those queues. In this case the tasks are on the same
computer or distributed (here Ada Distributed Annex can be of great
use). The deal is to balance the different tasks to avoid starvation
(from reader tasks) or waiting when queues are full.

At least in such architecture the number of tasks is mostly static. We
do not waste too much time creating and finalizing tasks/threads.

Anyway, here Ada concurrency and distributed support can be used instead
of the OpenMP and MPI couple as used in many Fortran and C/C++ applications.

Pascal.

-- 

--|------------------------------------------------------
--| Pascal Obry                           Team-Ada Member
--| 45, rue Gabriel Peri - 78114 Magny Les Hameaux FRANCE
--|------------------------------------------------------
--|              http://www.obry.net
--| "The best way to travel is by means of imagination"
--|
--| gpg --keyserver wwwkeys.pgp.net --recv-key C1082595