Cluster Computing And Visualization On The MASSIVE Grid Environment

MASSIVE (Multidisciplinary Applications-Oriented Simulation and Visualization Environment) is an integrated grid environment for engineering and scientific computation. A grid platform is a part of the MASSIVE environment, and it is designed oriented to large-scale numerical simulations, of which issues are related to computing power, storage, and network. The grid platform visually performs resource discovering, task scheduling, and result processing. It is built on the top of Globus Toolkit, and consists of two parts: a GUI and a grid server.The GUI has three modules: job manager, message service (MEMDS) and file transfer (MFTP). The job manager performs resources discovering, resources scheduling, task executing, and task monitoring. The MEMDS provides resource discovery with a user-friendly GUI, and the MFTP features large-scale data transfer with a user-friendly and interactive GUI.This thesis focuses on the grid server, which includes mainly execution of grid parallel tasks and management of parallel tasks on a PC cluster. As all nodes in a cluster have only local IP addresses, a parallel task cannot be executed smoothly. This thesis proposes and implements a solution for extending the MPICH-G2, and implemented it. The solution includes modifying the MPICH-G2 source files, coding a proxy program and supporting task management. The management of parallel tasks on a cluster includes resource allocation, process creation and process monitoring, which aim to load balance between nodes in the cluster, remote process creation and remote process monitoring, respectively.The thesis also discusses parallel visualization -- collaboration technique of visualization and a visualization tool named ParaView. The original ParaView is not collaborative. The thesis describes two ways of collaborative visualization, and implements one of them, thus the ParaView could be run in a parallel mode on the grid.