Scaling MPI to short-memory MPPs such as BG/L
Proceedings of the 20th annual international conference on Supercomputing
Scalable memory registration for high performance networks using helper threads
Proceedings of the 8th ACM International Conference on Computing Frontiers
| Hi-index | 0.00 |
Scalability to a large number of processes is one of the weaknesses of current MPI implementations. Standard implementations scale to hundreds of nodes, but no beyond that. Performance is often more important than scalability and thus some assumptions about resources are taken that will not scale well. In this paper, we show that characteristics such as the size and the sender of MPI messages are very predictable (accuracy above 90%). Then, we present some examples where current MPI implementations on a large configurationwould not work well and how this predictability could be used to solve the scalability problem.