The big and the small: Parallel Computing is becoming pervasive – are we ready?
IBM Corporation
Abstract
We can no longer rely on computational speed up through increases in clock speed. In order to continue the speed ups users have come to expect from computer companies the chip manufactures are turning to multi-core chips. Are we ready for such systems composed of multi-cores on a chip? Much of what we have learned on large distributed systems in principle will need to be applied to the multi-core efforts. I will compare a large system, Blue Gene, to a multi-core chip, the Cell BE. I will point out that lessons learn on systems like Blue Gene will help us with developing algorithms for Cell BE. The computational power may be accessible but most computational scientists are not prepared to fully take advantage of this. New approaches to computational problems need to be investigated, among these are multigrid methods. While progress is being made, there remain many challenges for the computational science community to apply ultra-scale systems and multi-core systems to “Big” science problems with impact on society that until now or in current implementations have fallen short of the mark. Finally, I will elaborate on opportunities that exist for the community to get access to Blue Gene and Cell BE systems.