Attended a couple tutorials today: Intro to OpenMP and Advanced Topics in OpenMP. Despite years in the trenches of parallel processing, I'd never studied OpenMP. Very code-at-the-traces stuff, not too useful for anything Perlish, but stimulated the neurons just the same. I'm not too keen on all the #pragmas and compiled-in stuff, but I suppose serious computational hackers need that.
It did give me notions about trying to hack Perl's iterator opcodes to provide auto-threading versions...guess I'll need to paw thru the compile sources, and try the midnight opcode swap hacks I've seen posted here and elsewhere
They gave us a flash drive w/ *all* the tutorial presentations on it, so I'll dive into the MPI stuff ASAP; maybe that'll be a bit more Perl-friendly (there are a couple MPI modules on CPAN, one recently updated).
Alas, my schedule this week won't permit me to attend the regular sessions, but they're mostly paradoxically brief: 30 minutes to discuss anything related to HPC seems a bit foolish, except for cocktail parties and Nova snippets.
However, I'm definitely going to the exhibit hall Wed. or Thursday. I was astounded at the number of exhibits/vendors, and the elaborate displays I caught a glimpse of from the convention hall lobby. I can only assume that there's serious $$$ in HPC these days - I'm guessing most of it is gummint money. Still, it does warm the heart of an old parallel coder to see new life in the discipline. However, based on overheard conversations, and the content of tech sessions and tutorials, I'm guessing the HPC hackers are still grinding out code the same way we did 15-20 years ago, which is a bit disappointing.
I'm still convinced there's a big opportunity for dynamic languages in this space, tho likely not for the serious computational stuff. And it resurrected a memory of something I'd been meaning to track down for a Thread::Sociable storage manager: RDMA.
