Improving the Performance of Grid-Enabled MPI by Intelligent Message Compression

In the Grid computing, one of the most important issues related to the Message Passing Interface (MPI) is to guarantee the expected performance in the wide area networks. To achieve high performance of MPI based communications, low latency, high bandwidth, effective communication mechanism, and effective management of processes are required. However, a long communication latency in the wide area network is an unsolved problem and is even difficult to reduce in the Grid environment. To solve those problems, this paper suggests an ApplicationSpecific Intelligent message Compression (ASIC) that can intelligently compress messages at the level of TCP, considering the current status parameters, i.e, the types of the MPI applications, the location information of processes, the network bandwidth, and the compression ratio. MPICH-G2 with ASIC can achieve better performance gain in MPI applications by around 125% than MPICH-G2.

• Type: Conference
• Year: 2003
• Where: IC
• Authors: Hwang-Jik Lee, Kyung-Lang Park, Kwang-Won Koh, Oh-Young Kwon, Hyoung-Woo Park, Shin-Dug Kim
• Keywords: IC 2003, IC 2007, MPI Applications, MPI Based Communications, Wide Area Network