Boris Jeremic wrote:
> - regarding parallel inplementation, I think it would be wise to use
> MPI as it is widely available. I have a small Beowulf cluster runing
> LINUX and MPI (or PVM) is the easiest and most portable way to go (the
> same applies to large national parallel machines which seem to have MPI
> or PVM ready). I know Dirk from Boulder (although he might not remember
> me since I was only a student when we discussed some things) and I know
> that they (CS at Boulder) have all the big machines they need
> (Paragon, SP/2, SGI Onyx . . . ) so They might not be interested in MPI
> so much. Anyway if Dirk is interested we can only benefit from his
> knowledge and then somebody else might take it to MPI,
I don't know a lot about it, but apparently MPI is not efficient
for shared memory/SMPs; threads are better in this situation. One
advantage of SMARTS/TULIP is that it can use threads for communication
between CPUS in an SMP, and MPI for between boxes.
One of the ACL folks is currently implementing TULIP on top of
MPI, which might solve portability concerns.
Cheers,
Todd
--------------------- blitz-dev list --------------------------------
* To subscribe/unsubscribe: mail to majordomo@oonumerics.org, with
"subscribe blitz-dev" or "unsubscribe blitz-dev" in the body of the message
* Blitz++ web page: http://oonumerics.org/blitz/
This archive was generated by hypermail 2b29 : Wed Feb 20 2002 - 04:30:04 EST