Re: [Boost-users] hybrid parallelism

Subject: Re: [Boost-users] hybrid parallelism
From: David Abrahams (dave_at_[hidden])
Date: 2010-11-03 22:09:27

At Thu, 4 Nov 2010 00:19:58 -0000,
Hicham Mouline wrote:
> 
> > From: boost-users-bounces_at_[hidden] [mailto:boost-users-
> > bounces_at_[hidden]] On Behalf Of Dave Abrahams
> > 
> > If I were you I would start by trying to do this with N processes per
> > machine, rather than N threads, since you need the MPI communication
> > anyway.
> 
> Just temporarily? You would still after that add a layer of multithreading
> to each process, and have only 1 process per machine, after that, no?

Not necessarily.  It would depend if the performance was good enough
or not.

> A 1 process N threads in 1 machine is probably better total wall time than
> just N mono threaded processes because of the no need to duplicate the input
> memory to the tasks.

You can use shared memory for that, if you want.  Whether there is a
tangible advantage probably depends on the size of your input.

> The question I really wanted to ask about is that I expect to have M*N
> outstanding threads (M computers, N threads in each process) just sitting
> there waiting for jobs.
> Then from the user interface, I click and that starts 100000 tasks, then it
> is spread all over the M machines and N threads in each process. Then result
> comes back, displayed...
> Then user clicks again and same thing happens.
> 
> You're saying this is doable with Boost.MPI + MPI impl?

If you are determined to have threads, you'll need more than just MPI,
but yes, you can use Boost.MPI to handle the communication and
synchronization across machines.

> I wasn't expecting to divide the tasks into finer grained ones. All the
> tasks are atomic and have about the same exec time. It's rather pass
> 100000/M tasks to each machine, then divide this number by N for each thread
> in that process. This last bit is up to me to code.
> Ideally, the task is just a functor with operator() member and M machines
> and N threads are treated similarly. I guess it's up to me to write some
> abstraction layer to view the whole M*N in a flat way.

That's why I'd start with processes and MPI; it already has the
abstraction layers that make two processes running on the same machine
look identical to the same two processes running on different
machines.

-- 
Dave Abrahams
BoostPro Computing
http://www.boostpro.com