> The paper says:
> "We are working on a distributed implementation of Concurrent Haskell.
> Once nice property of MVars is that they seem relatively easy to implement
> in a distributed setting..."
> I assume that they are not referring to GPH here.
> (I was surprised that at this statement given what I presume are the
> substantial difficulties of defining a wire format for lazy
> datastructures).
>
> I would assume that the distributed implementation gives the programmer
> some access to whether a machine is "up" or I suppose that you could use
> distributed mvars to design your own ping (cool!)
OK, here's the current story.
* Concurrent Haskell works on uniprocessors only. It's intended for
programs that need explicitly-forked, concurrent, I/O-performing
threads. Processes are explicitly forked with forkIO, which is
in the IO monad.
Results are necessarily non-deterministic -- that's part
of the point! There is no attempt to use parallelism to gain
performance.
* GpH (Glasgow Parallel Haskell) works on a variety of multiprocessors.
It's specifically intended to harness parallelism to gain performance.
Parallelism is explicitly sparked with `par`, but the RTS is free to
ignore such sparks. `par` is not in the IO monad: it has type
par :: a -> b -> b
Results remain completely deterministic even on a multiprocessor.
I'd like to implement a sort of integrated system. Certainly it
would be nice to be able to make Concurrent Haskell into Distributed
Haskell. To to this MVars would need to work across processors
(not hard), and forkIO would need a variant that said "fork this
process on this other host". The goal would still not be performance-through-
parallelism; rather, it is part of the *specification* of a distributed
program that it different parts should execute in different places.
Perhaps one could also have some combination of DH and GpH; after
all they use similar underlying mechanisms.
I believe that Kevin Hammond, Phil Trinder and Rita Loogen are planning
to do something like this, but I'm afraid it's not directly in my
sights!
Einar Karlsson ([EMAIL PROTECTED]) has been doing a
huge application in Concurrent Haskell so he may have other comments.
I hope that helps to clarify where things are at from the implementation
point of view. One way to get us hackers to do things is to
say loud and clear what would actually be useful to you.
Best wishes
Simon
