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

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

Am Freitag, den 25.03.2011, 10:55 +0100 schrieb Alain Frisch:
> Hi Fabrice,
> 
> On 03/25/2011 12:13 AM, Fabrice Le Fessant wrote:
> >    Actually, I had a discussion two weeks ago with Xavier and Damien
> > about this issue. There is some kind of agreement that the ocaml way of
> > supporting multicore would be to have several runtimes running in the
> > same process, in different threads. That way, the GC would still be
> > mono-threaded, so almost no speed loss for mono-threaded programs (i.e.
> > currently all OCaml programs ;-) ). There would be some kind of "fork"
> > function, that would create a new thread running a function in a new
> > heap, probably generated by a copy-on-need algorithm. The different
> > threads would not share heap memory, but would be allowed to share
> > structures outside of their heaps, probably for simple types like
> > strings and int/float arrays (or using the Ancient library).
> 
> This sound very reasonable, and it's good to know that the subject is 
> being discussed.
> 
> What are the advantages of this approach over multi-processing (+ 
> explicit shared memory)?  Multi-processing brings extra isolation and 
> safety; and it requires no change to the runtime system or the FFI.
> 
> One advantage of the approach you describe is that it would allow to use 
> native libraries that support multi-threading. Do you see other advantages?

It is an interesting approach. There is at least one big advantage: When
new memory is allocated that is to be shared by all workers, it is
mapped at the same address. In the multi-process design this is
difficult to organize - there is no system call for this. The only safe
way to get this effect is to allocate memory before new workers are
forked (the unsafe way is to manage addresses manually, but this is
quickly becoming very OS-dependent).

Being able to map shared memory at the same address is important for
some uses, especially when ocaml values are moved there directly in the
style of Ancient. For my Camlbox data structure I had to work around
this issue (and it was feasible), but not all interesting data
structures would allow that.

Gerd

> Cheers,
> 
> Alain
> 


-- 
------------------------------------------------------------
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
------------------------------------------------------------