[Gomp-discuss] parallell vs. distributed

[Lars Segerlund]

Scott Robert Ladd wrote:
> Lars Segerlund wrote
...[snip]...
> >  What about the distinction of 'parallell' and 'distributed' regions,
> > does everyone think this is a good idea ?
>
>
> I'm not certain what you mean here. When I think of "distributed", I think
> of an application like address@hidden, where work is shared across a loose
> network of heterogenous machines. The OpenMP model is vetry much based on
> "threads" (however a given system defines those) running in a shared-memory,
> multiple processor environment.
>
> ..Scott

 The reason I want to make a distinction is that fx. odinmp has an 
implementation which can run ontop of mpi ! ( hence distributed ), and 
thus is not sharing memory with the other 'threads'.

 The classic approach has been to generate subroutines for the 
parallell parts, which copy in and out all needed parameters ( perhaps 
some syncronization and so on but you get the general idea ), this 
overhead is circumvented by fx. intels compiler assuming shared memory, 
and only generating parallell sections, ( of course with a set of 
private variables for each thread ), and the shared memory is really shared.

 Right now we might not have too much use of the 'distributed section' 
attribute, but it would make it easy for us to do future enhancements, 
or code as subroutines for profiling or diagnostics.

 In this context the parallell sections and distributed sections would 
work from each end, as a distributed computation is best done on a 
larger data set, and parallell optimizations ( thread ) would in general 
work best for a bottom up approach.

 It's a bit of future enhancement/safety that I thought would be nice :-) .

 I think most people didn't like the idea, so I gave it up, I still 
think it might be nice if one would like to run with mpi or pvm support, 
as a lot of existing clusters could use this feature.

 / regards, Lars Segerlund.