HPC Blog Dr DK Panda is a world-recognised expert on parallel programming and networking. He's a Distinguished Scholar at The Ohio State University and his research group has developed the MVAPICH2 (high performance MPI and MIP+PGAS) libraries for InfiniBand, iWARP, and RoCE with support for GPUs, Xeon Phi, and virtualization.
His resumé and list of accomplishments goes on and on, but take it from me, he's the man when it comes to programming models and networks.
He recently delivered a talk at the HPC Advisory Council Lugano Conference in Switzerland, which covered pretty much the entire range of challenges in designing programming models and environments for Exascale systems.
The challenges are huge. We're talking about systems with millions of processors and billions of cores, plus massive numbers of accelerators of various types.
Systems at this scale will have to support highly efficient internode and intranode communications and provide scalable collective operations with a non-blocking network. Furthermore, they will require a topology and energy-aware design that can optimise for message speed but also energy usage.
Fortunately, his talk at Lugano was recorded and you can get the slides here and see the video here. Dr Panda dives deep and is very thorough in his presentation – so you'll need to be wearing your thinking caps when watching.