Re: non-blocking ops vs. threads

Al Geist (geist@msr.EPM.ORNL.GOV)
Fri, 7 Jul 1995 22:35:44 -0400

>From: Tony Skjellum
>we should identify properties
>in addition to POSIX requirements that would make MPI happy
>with threads, and then push for such a "High Performance POSIX
>Threads" option to be supported by vendors.

While this would be nice, my concern is that this may delay
the release of MPI-2 by vendors if they have to create a
High Performance POSIX Threads package for their machine first.

My feeling is that threads is the way to go in the long run,
but the present environment does not lend itself to portability.
We just got through testing a multithreaded version of PVM.
Porting to just a few machines was an awful mess.

And what about portability to non-UNIX boxes like microkernel compute nodes
or (gag me) Windows NT?

My e cents worth,
Al Geist