[Gllug] Reia

[Richard Jones]

On Sat, Feb 21, 2009 at 12:37:27AM +0000, John Hearns wrote:
> 2009/2/20 James Laver <gllug at jameslaver.com>:
> > On Fri, February 20, 2009 5:40 pm, - Tethys wrote:
> >> Does each scheduler have a pool of
> >> processes for which it's responsible? Can a process move between
> >> schedulers (and hence cores)?
> >
> > Yes, including moving processes over a network.
> 
> A reference to this would be interesting.
> As someone who has looked after a Mosix cluster (in the basement of
> the Natural History Museum) I would like to see what the requirements
> for this are - eg. identical kernels etc.

Erlang is compiled to bytecode which is interpreted in each node, so I
imagine that it's not as hard to migrate as real processes, but is
also the reason why single thread Erlang performance is slow[1].

There is/was this project to replace the single threaded code with
OCaml which would obviously be super-fast while maintaining the safety
features of functional languages:

  http://code.google.com/p/erlocaml/

although it doesn't seem to be going anywhere ...

OCaml has its own distribution systems anyway.  Probably the most
widely used is plain MPI, which supports serializing native OCaml
objects between nodes (but NOT migration of processes):

  http://pauillac.inria.fr/~xleroy/software.html#ocamlmpi

Rich.

[1] When I say "slow" I do of course mean the same sucky speed as
other interpreted languages like BASIC, Perl, Python and Ruby.


-- 
Richard Jones
Red Hat
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug