Grid is an attractive environment that provides numerous available resources by high speed network, and its utilization can be a solution of grand challenge problems. There are several implementations that provide Grid environment. Globus is such a middleware, which solves Grid problem defined as controlled and coordinated resource sharing and resource use in dynamic and scalable virtual organization. In order to perform parallel jobs in Grid a useful interface is needed. MPI(Message Passing Interface) is standard of parallel mechanism used for a long time. MPICH-G2 is a grid-enabled MPI based on the MPICH library, and it makes an efficient use of underlying Grid. Many application using MPICH-G2 are developed in Grid. A communication performance is one of major issues in Grid environment connected to hetero-geneous network. Collective communication operations provide a useful interface of group communication and consist of point-to-point operations of MPICH-G2. Scheduling collective communication primitives of MPICH-G2 underlying Grid is a main factor affecting the communication performance of MPICH-G2.
In this thesis, we propose a packet-level parallel data transfer and a two-phase scheduling algorithm for collective communication primitives in order to improve the performance of MPICH-G2.