[Haskell-cafe] Bulk Synchronous Parallel

[Jason Dusek]

2010/4/21 Aaron D. Ball <aarondball at gmail.com>:
> I don't need a tool that automatically figures out how to distribute
> any workload in an intelligent way and handles all the communication
> for me.

  You are right in general. Only if you want to rely on purity and a
  few source code annotations to get you parallelism relatively
  cheaply do you care about these compiler approaches. This is something
  that Haskell can do that Ruby, C and friends really can not do -- thus
  I mention it.

> If I have the basic building block, which is the ability to
> serialize a Haskell expression with its dependencies and read them
> into another Haskell instance where I can evaluate them, I can handle
> the other pieces, which are
>
> - passing strings back and forth in whatever way is convenient
> - deciding how to divide up my workload.

  Do add also, configuring servers and their connections.

> In the Ruby universe, DRb combines the serialization and "passing
> strings around" job and lets me figure out how to divide up the work,
> and it would be delightful if there were something similarly simple in
> the Haskell world.

  I think Holumbus has got some promising stuff for user-managed
  distributed workers:

    http://holumbus.fh-wedel.de/trac/browser/distribution

  What do you think?

--
Jason Dusek