Research On Grid Parallel Implementation Of The FDTD Algorithm Using Globus Toolkit

Currently, parallel implementation of FDTD algorithm is changing from the original MPP (Massive Parallel Processors) computing to network parallel computing (NPC), which is a preference by using high performance processors and high speed network equipments. For NPC, programming model, executing environment and development tools are crucial in terms of computing capacity. Meanwhile, grid computing is normative parallel distributed computing by combining the resources of computing, storage and communication out of a collection of different, heterogeneous and geographically separated PCs. The goal is to create a simple but large and powerful self-managing virtual computer. Profound researches on parallel implementation of FDTD have been made in recent years, accompanying with the research on MPI programming model and MPICH-G2 grid parallel development tool. These researches contribute a lot to the development of grid parallel computing.This dissertation designed and implemented a shared-resources and cooperative grid parallel system, which is a grid-enabled MPI environment that can run MPI FDTD program successfully. Additionally, we set up a service model and optimized the MPI parallel program.This dissertation focused on the implementation of a MPICH-G2 grid parallel system based on Globus Toolkit, including the installation of Globus Toolkit and the construction of distributed certification. Then we set up the Globus service model, which encapsulating a series of services such as grid environment, job submission, allocation, execution and supervision. This service model is in accord with Globus's service-oriented trend, and present detailed analysis of the execution process of parallel programs on grid system by combining grid architecture. This was followed by a strategy of FDTD parallel computing in light of SPMD parallel architecture and MPI programming model. Additionally, the optimized program was presented by using the parallel sockets provided by Globus's GridFTP.Finally, investigation about the influences of parallel sockets on MPIDJ-G2 based on two experiments is given, further more, an important conclusion is made: grid computing dispenses with expensive computer equipment software and hardware investment, and if provided with a certain network bandwidth, low-cost grid parallel computing is more preferred in wider area when comparing with other high performance computing methods.