This paper describes the basic concepts of our solution to improve the performance of Ethernet Communication on a Linux Cluster environment by introducing Reliable Low Latency Ethernet Sockets. We show that about 25% of the socket latency can be saved by using our simplified protocol. Especially, we put emphasis on demonstrating that this performance benefit is able to speed up the MPI level communication. Therefore we have developed a new BTL component for Open MPI, an open source MPI-2 implementation which offers with its Modular Component Architecture a nearly ideal environment to implement our changes. Microbenchmarks of MPI collective and Point-to-Point operations were performed. We see a performance improvement of 8% to 16% for LU and SP implementations of the NAS parallel benchmark suite which spends a significant amount of time in the MPI. Practical application tests with Abinit, an electronic structure calculation program, show that the runtime of be nearly halved on a 4 node system. Thus we show evidence that our new Ethernet communication protocol is able to increase the speedup of parallel applications considerably.

This paper describes the basic concepts of our solution to improve the performance of Ethernet Communication on a Linux Cluster environment by introducing Reliable Low Latency Ethernet Sockets. We show that about 25% of the socket latency can be saved by using our simplified protocol. Especially, we put emphasis on demonstrating that this performance benefit is able to speed up the MPI level communication. Therefore we have developed a new BTL component for Open MPI, an open source MPI-2 implementation which offers with its Modular Component Architecture a nearly ideal environment to implement our changes. Microbenchmarks of MPI collective and Point-to-Point operations were performed. We see a performance improvement of 8% to 16% for LU and SP implementations of the NAS parallel benchmark suite which spends a significant amount of time in the MPI. Practical application tests with Abinit, an electronic structure calculation program, show that the runtime of be nearly halved on a 4 node system. Thus we show evidence that our new Ethernet communication protocol is able to increase the speedup of parallel applications considerably.