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

["Jon Harrop" <jon@ffconsultancy.com>]

Alexy wrote:
> I'm not claiming that something is correct or not -- I'm just
> saying that replacing map by pmap is easy, while rewriting
> in MPI style is complex.

Complex not only because you must pass information around by hand but also
(in the absence of a shared heap, e.g. in OCaml) because you must deep copy
your data structures in order to send them which turns O(1)
pass-by-reference into O(n) pass-by-value and that, in turn, makes the
performance of MPI code much harder to predict.

> This is an obsolete urban legend.  JVM has the most mature
> GC out there and computational performance often on par
> with C when loaded and running.

The JVM does have a very evolved GC and it can often yield competitive
performance but it also has its fair share of deficiencies that can make it
much harder than necessary to attain competitive performance. In this
context, type erasure for generics and the lack of value types are major
issues. These culminate in generic code doing far more allocation than
necessary. Although this is partially offset by the JVM's GC there are still
important situations (e.g. generic hash table) where the JVM is many times
slower than alternatives like .NET (which uses reified generics and has
value types):

  http://fsharpnews.blogspot.com/2010/05/java-vs-f.html

> In my social networking benchmark...

That's a very interesting benchmark!

> Clojure's performance improves by leaps and bounds, e.g.
> using primitives as efficiently as in Java, and I think OCaml
> would benefit from a similar set of primitives --

Absolutely. There's no theoretical reason why this functionality could not
be available from OCaml too but it is a major project and, as I said, nobody
has enough motive to undertake it.

> then it would be the most practical ML-style FP language,
> the prize now in fact held by Scala.

I'd like to see F# too. :-)

Cheers,
Jon.