Re: [Caml-list] Efficient OCaml multicore -- roadmap?

[Gerd Stolpmann <info@gerd-stolpmann.de>]

Am Freitag, den 22.04.2011, 10:06 +0200 schrieb Fabrice Le Fessant:
> On 04/21/2011 09:03 PM, Christophe TROESTLER wrote:
> > For the initial question: If OCaml gets proper parallelism, I believe
> > it is good not to neglect the needs of scientific users (to which I
> > belong) ― maybe they cannot be 100% met due to other considerations
> > but rejecting them from the start does not feel right to me.
> 
> I think nobody is rejecting the needs of scientific users. OCamlPro
> actually started a long term project called "numcaml", to provide
> something similar (including syntactically) to numpy, which, I think,
> turned a lot of scientific users into Python fans. We are also thinking
> about how to address fine-grain parallel numerical computations using
> our current multicore solution.
> 
> My point was just that J.H. has been writting everywhere, in every
> single forum in the world, that OCaml is bad because it could not
> address fine-grain parallelism, for his personal needs. I thought it was
> time for the silent majority of satisfied OCaml users to say, here and
> in those forums, that OCaml is actually addressing their most important
> concerns.

That would be good.

I've just released an example of using Netmulticore for fine-grained
parallelism:

http://blog.camlcity.org/blog/multicore3.html

J.H. is certainly right that he demands shared memory. Certain
algorithms cannot be done without. But shm is possible, even with
current ocaml - and Netmulticore demonstrates it. AFAIK the other
multi-process environments lack it.

Gerd

> Best,
> Fabrice
> 


-- 
------------------------------------------------------------
Gerd Stolpmann, Bad Nauheimer Str.3, 64289 Darmstadt,Germany 
gerd@gerd-stolpmann.de          http://www.gerd-stolpmann.de
Phone: +49-6151-153855                  Fax: +49-6151-997714
------------------------------------------------------------