Re: Parallel octave ??

[Javier Fernández]

Michael Creel wrote:
> About mpirun being the way to go, I think that I agree in most cases.
> But the on-the-fly spawning works well, too, and for some of my work I
> can't really figure out how to easily adapt it to the mpirun model
> without loosing a lot of flexibility.
> Cheers, Michael
>   

Hmm, you are right, I was concentrated on escaping from parfor and I 
didn't notice I was getting too sententious, pompous, and all those bad 
"...ous" :-) sorry for that, I take it back. I have been thinking on how 
to reword it, but it's better if I take it directly back ;-) Anybody can 
go their way, which is obviously true.

Also, prospects are good since the OMPI Team not only was aware of the 
spawn "problem" in the 1.2.x series, but they had solved it in the 1.3.x 
series (even before anybody mentioned it was not working the way they 
wanted or were used to), so it is most probable that the 1.4.x series' 
"..._spawn" will work as well as LAM's used to do. I think chances are 
good that the new OMPI series is downloadable before the LAM downloads 
dissappear (if they ever do :-). I mean, when the 1.4.x is out, you'll 
be able to move to OMPI without any problems.

I lost track of the presentations in the HPC Challenge awards 
(http://www.hpcchallenge.org/), and it's difficult to extract 
conclusions from brief comments in slides, but it's noticeable that the 
only appearance of "parfor" in Moler's sc2005 slides 
(http://www.hpcchallenge.org/presentations/sc2005/Moler.pdf) is in an 
enumeration (p.3/28)... no parfor appears in the code fragments, and he 
is clear in the sentence "Think matrices..." Hmpf, I'm again trying to 
escape from parfor, providing excuses and excuses... Sorry again :-)

-javier