On Mon, Jun 12, 2006 at 09:36:23PM -0400, Joe Landman wrote: > This frustrates software builders, and end users. My point is that > there is a better way, and Greg indicated that he had supported/proposed > it.
By the way, chatting with other interconnect vendors and also with ISVs, we're pretty much all for it. The group that hated the idea was MPI implementors, both free and commercial. I tried several different forms of sweet-talking, none got anywhere. It's clearly a sign that someone else needs to do the sweet-talking! Also by the by, what Microsoft is doing isn't really an ABI, but it acts like one: you use MPICH-2's header files and replace the MPICH-2 DLL. Now for interconnect vendors this is easiest to do if you have an MPICH-2 adi3 device. You'd have to be mildly masocistic to write a layer that translates to some other MPI guts, but if anyone does, please share. > Hmmm.... most of the apps I have seen are software bound at some point. > Some scale really well, but those are rare. 16-32 way runs are fairly > typical at customer sites. A few do more, many do less. One thing that interests me about the MPI market is that people's experiences are so different, depending on what kinds of code and what size problems they run. But, do keep in mind that a better interconnect will usually allow you to scale to more cpus. -- greg _______________________________________________ Beowulf mailing list, Beowulf@beowulf.org To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
