Analyzing the impact of communication imbalance in high-speed networks

Abstract

In this work we analyze the communication load imbalance generated by irregular-data applications running in a multi-node cluster. Experimental approaches to diminish communication load imbalance are evaluated using a hybrid programming model MPI+OpenMP including certain optimizations like computation-communication overlap, issuing communications in parallel and a new proposal based on message fragmentation in order to take advantage of the eager-protocol. Performance results show that overlapped versions can obtain a great benefit of this optimization because it avoids switching to rendez-vous protocols. However, non-overlapped versions showed better performance than overlapped ones. To evaluate also the impact due to network latency, the work has been tested on two high-speed interconnection networks: Infiniband and 10Gigabit Ethernet. In this case, the optimizations in the non-overlapped miniFE benchmark reached and improvement up to 7% on Infiniband and 11% on 10Gigabit.