The check-pointed and error-recoverable MPI java library of agent teamwork grid computing middleware
Munehiro Fukuda, Zhiji Huang · 2005
We are implementing a fault-tolerant mpiJava API on top of the AgentTeamwork grid-computing middleware system. Our mpiJava implementation consists of the mpiJava API, the GridTcp socket library, and the user program wrapper, each providing a user with the standard mpiJava functions, facilitating message-recording/error-recovering socket connections, and monitoring a user process. This paper presents the application framework, mpiJava implementation, and communication performance in AgentTeamwork.