Pedro Gonnet
Durham

High-Performance Computing and the need for good Graph Algorithms

For the past five years, computers have become faster only by becoming
more parallel: Any standard desktop PC now has at least four cores and a
GPU with a hundred more. In order to exploit this new massive
shard-memory parallelism for High-Performance Computing applications,
new algorithms and/or programming approaches are needed. In this talk, I
will present some parallelization methods and show how, in general, an
efficient parallelization relies on good graph partitioning algorithms.
The purpose of this talk is not to present any results, but to hopefully
ask the right questions.