-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1



On Sun, Dec 13, 2009 at 2:26 AM, v01d  wrote:

> Hi,
> I'm a little interested in a possible future where Ruby could be used
> for heavy processing stuff (but by means of calling computationally
> heavy C code, and take advantage of multi-core processors by doing
> several of these calls in parallel native threads) and I just found
> out about Ruby's "rb_thread_blocking_region" C call.
>
> What came to my mind is that maybe FFI could be used here to make
> writing these type of parallelized Ruby extensions easier, by just
> doing something like:
>
> module RubyAPI
>  extend FFI::Library
>  attach_function :rb_thread_blocking_region,
> [ :pointer, :pointer, :pointer ], :void
> end
>
> then maybe I'm wrapping a HeavyComputationLibrary:
>

My 2c:
Context: I take 'HeavyComputation' to mean 'saturates more than one machine'.
In that case I think you'd be using somthing like this:
http://github.com/abedra/mpi-ruby

ffi-mpi-ruby would be interesting.

My preference has been to _try_ avoid threading and any issues it
causes.  I would rather just throw more instances of a ruby script
(processes, somtimes processes>=processors/cores) at the problem -
this does scale beyond one machine.  Threading doesn't.
It has been a while since I used MPI and there may be threading buried
in the library above.

Of course 'HeavyComputation' might mean using only one machine.  Then
scaling is not an issue and maybe threading is undisputably the best
way to go?
As I said, just my 2c.

HH

> module HeavyComputationLibrary
>  extend FFI::Library
>  attach_function :long_running_computation, [ :pointer ], :value
>  attach_function :unblock_computation, [ :pointer ], :void
> end
>
> and then do something like:
> RubyAPI.rb_thread_blocking_region
> (HeavyComputationLibrary.long_running_computation, data_ptr,
> HeavyComputationLibrary.unblock_computation)
>
> Obviously there are a couple of non-existing things here (like passing
> the function pointer of a C function (maybe there's already a way to
> do that?), usage of :value), but I wanted to ask if this is feasible
> or not.
> I think that something like this could ease writing parallel-aware
> Ruby extensions (maybe with some help of Ruby Inline) for existing C
> libraries. Then again, I may be wrong.
>
> Matt
>
> PS: I'm obviously talking about Ruby 1.9 and the function call defined
> here: http://blade.nagaokaut.ac.jp/cgi-bin/scat.rb/ruby/ruby-core/10252
>
>

- --
πόλλ' οἶδ ἀλώπηξ, ἀλλ' ἐχῖνος ἓν μέγα
[The fox knows many things, but the hedgehog knows one big thing.]
  Archilochus, Greek poet (c. 680 BC – c. 645 BC)
http://wiki.hedgehogshiatus.com


-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.9 (GNU/Linux)
Comment: Use GnuPG with Firefox : http://getfiregpg.org (Version: 0.7.10)

iEYEARECAAYFAksj8EYACgkQL/DTCig+NhfFWQCeKK5ohPsJzjieM2y1HSv7l4Dq
8gIAn1b97tMd/o4sBg1uPX+JJFRVDMf7
=Mj85
-----END PGP SIGNATURE-----