3 ms·
I don't think it's because MPI is better, it's just most of the supercomputers I have access to require use of MPI like constructs.
by noobermin 8y ago
I don't think it's because MPI is better, it's just most of the supercomputers I have access to require use of MPI like constructs.
- cbkeller 8y agoBetter than? Not necessarily disagreeing, but I'm not sure what the alternatives even are at the same level of abstraction. I mean, there's PNNL's global arrays [1] but that's higher level, or Sandia Portals [2] which is lower/transport level.. Perhaps there are newer/alternative options I don't know about? [1] http://hpc.pnl.gov/globalarrays/ http://hpc.pnl.gov/globalarrays/ [2] http://www.cs.sandia.gov/Portals/portals4-libs.html http://www.cs.sandia.gov/Portals/portals4-libs.html
- gnufx 8y agoGlobal arrays is normally used over MPI anyhow. I guess there's SHMEM, but that's integrated with at least OpenMPI (and others, I think). CHARM++ has been used at scale, but it's semi-proprietary.
- timClicks 8y agoAnd that's often because network hardware understands MPI and is able to optimize flows between nodes at far lower latency than TCP.
- Something1234 8y agoThat's really cool. Source?
- timClicks 8y agoI used to work in HPC. The Mellanox gear, specifically InfiniBand is very good. Fun fact: if you're working at a Saudi Arabian HPC center, say KAUST, your interconnects are purely Ethernet. Mellanox is (partially?) an Israeli company, and that's not very politically comfortable with procurement.