Is it time to start rethinking concurrency in OCaml?

I have followed the argumentation of only using one native thread for
the OCaml runtime.
I can easily see how this can increase performance and simplify implementation.
I can also see that spawning new processes makes sense, so you get a
local heap for each task.

However, as we move forward it seems that we will get more than a few
cores on the same computational node according to the following
article:

Intel says to prepare for 'thousands of cores':
http://news.cnet.com/8301-13924_3-9981760-64.html?part=rss&subj=news&tag=2547-1_3-0-20

As I see it, it is not feasible to spawn a new process with a local
heap for each core, when the number of cores increases dramatically.

I am not sure that a parallel GC is a sufficient solution either due
to the high contention on memory, at least unless it provide some
additional core affinity features. I believe some level of compiler
support is needed in the not so distant future such that enough
primitives are available to build powerful multi-core aware libraries.
One approach could be micro heaps with core affinity and handle
mutable memory specially.


Regards,
Mikkel

_______________________________________________
Caml-list mailing list. Subscription management:
http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
Archives: http://caml.inria.fr
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
Bug reports: http://caml.inria.fr/bin/caml-bugs

Reply via email to