Over at the ISC Blog, Dr Mirko Rahn from Frauenhofer writes that Partitioned Global Address Space (PGAS) approaches have become a hot topic for exascale computing. While there is much work to be done in this area, the EC-funded EPiGRAM project has identified the gaps to be filled when attempting to master the Exascale challenge with PGAS.
Altogether, one can choose between a non-scaling implementation that has low latency and a scaling implementation with high latency. This seems to be the case for each and every global property. And also the solution to trade latency for memory seems to be a generic one. There are two solutions to this theoretical limitation. First, one could imagine a system that does not have names for each entity. It is not clear whether or not such a system could be implemented for all the important applications. Second, one could write applications that can tolerate high latencies. In general these kind of theoretical limitations are not gaps in the programming model but they become important issues when the number of components grows to exascale.
Read the Full Story * Learn more about the EPiGRAM project in a special BoF session at ISC’14 * Register for ISC’14