Speaker
Project(s) or EGEE activity presenting the demo or poster (project or activity names only)
NA4 regional participation for UK, Ireland, France
Abstract
As the HPC community moves beyond the Petaflop scale, the rest of us are now immersed into the multi-core and many-core era. Increases in network bandwith and memory is bluring the distinction between typical HPC and non HPC resources. The developement of scalable and portable applications on these new architectures is still lacking behind, for both the HPC community who suddenly have to program for 100s thousands of CPU cores, and the Grid community who has to deal with distributed clusters, of be it, smaller number of cores. Pragmatic approaches such as hybrid MPI, OpenMP or Pthreads have limitations, as each was designed with one paradigm in mind. A new breed of Partitioned Global Address Space (PGAS) languages and runtimes, that combine the benefits of message passing (MPI) with shared memory (OpenMP) programming, have the potential not only to ease application development, but to allow portability and scaling between GRIDs and HPC centres.
Special requirements other than the set up mentioned in the CfA text.
none
