Evaluating Communication Performance In Bluegene/Q And Cray Xe6 Supercomputers

Summary form only given. Evaluation of the communication performance in supercomputers Blue Gene/Q and Cray XE6 using MPI and lower level libraries are presented. MPI is widely used for to its simplicity, portability and straightforwardness, however it also introduces overhead that degrades communication performance. Some applications with performance constraints cannot tolerate that degradation. Recent supercomputers such as the Blue Gene/Q and CrayXE provide lower level communications libraries such as PAMI (Parallel Active Message Interface) and uGNI (User Generic Network Interface). Our experiments show that with different communications modes: one-sided, two-sided, inter-node, intra-node we can always achieve higher performance with certain message sizes. These results will enable us to develop a GLEAN- light-weight API, which can be called directly from applications without passing through MPI to reduce/remove overhead.