Threaded ocaml code, native threads, and linux SMP

Threaded ocaml code, native threads, and linux SMP

[Julian]

I wrote the following CPU-bound function in order to test the
Thread module in ocaml:

    let rec loop pair = match pair with
              (100000, 50000) -> ()
            | (     i, 50000) -> loop (i+1, 0)
            | (     i,     j) -> loop (i, j+1) ;;

This function was used in two test programs:

* Version 1 *

    let t1, t2 = Thread.create loop (0,0),
                 Thread.create loop (0,0) ;;

    Thread.join t1;;
    Thread.join t2;;


* Version 2 *

    loop (0,0);
    loop (0,0);;


The test programs were each compiled with the following command:
(ocaml 3.08, debian sid, linux kernel 2.6):

    ocamlopt -thread -o <name> unix.cmxa threads.cmxa <name>.ml


I then tried running the two versions on an SMP machine. I found that
whereas I expected version 1 to run roughly twice as fast, they actually
took the same amount of time.

Closer inspection using "ps" revealed that only one processor was being
used. When I wrote a similar test program in C, calling the pthread
functions directly, the threads were run on separate CPUs as expected.

The "ldd" command reveals that the C version and the ocaml version were
relying on the same native pthread library.

So why doesn't the ocaml version use 2 processors?
Is it a flaw with my program?
Did I compile the program incorrectly?


Thanks for any help you can give,
Julian

Re: [Caml-list] Threaded ocaml code, native threads, and linux SMP

[Kip Macy]

MT garbage collection is both complex and fairly high in overhead. Ocaml
only supports user level threads. If you want to use both processors you
need to use multiple processes.

		-Kip



On Fri, 17 Dec 2004, Julian wrote:

> I wrote the following CPU-bound function in order to test the
> Thread module in ocaml:
>
>     let rec loop pair = match pair with
>               (100000, 50000) -> ()
>             | (     i, 50000) -> loop (i+1, 0)
>             | (     i,     j) -> loop (i, j+1) ;;
>
> This function was used in two test programs:
>
> * Version 1 *
>
>     let t1, t2 = Thread.create loop (0,0),
>                  Thread.create loop (0,0) ;;
>
>     Thread.join t1;;
>     Thread.join t2;;
>
>
> * Version 2 *
>
>     loop (0,0);
>     loop (0,0);;
>
>
> The test programs were each compiled with the following command:
> (ocaml 3.08, debian sid, linux kernel 2.6):
>
>     ocamlopt -thread -o <name> unix.cmxa threads.cmxa <name>.ml
>
>
> I then tried running the two versions on an SMP machine. I found that
> whereas I expected version 1 to run roughly twice as fast, they actually
> took the same amount of time.
>
> Closer inspection using "ps" revealed that only one processor was being
> used. When I wrote a similar test program in C, calling the pthread
> functions directly, the threads were run on separate CPUs as expected.
>
> The "ldd" command reveals that the C version and the ocaml version were
> relying on the same native pthread library.
>
> So why doesn't the ocaml version use 2 processors?
> Is it a flaw with my program?
> Did I compile the program incorrectly?
>
>
> Thanks for any help you can give,
> Julian
>
> _______________________________________________
> 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
>