| The main objective of this research is to explore the computing power of a network of workstations for the distributed execution of computationally intensive programs and to evaluate the overall network performance for such applications. The dynamic scheduling mechanism developed for this purpose requires that an application program be represented as a collection of subprograms in a task graph format showing the subprograms' dependency. An application program, described in such a format, is scheduled for parallel execution in the network in accordance with the load status information of the workstations as follows. The independent subtasks in a particular level of the task graph are distributed among the idle or lightly loaded processors (slaves) in the network by the user workstation (master) for parallel execution. The partial results are then collected by the master to ensure the synchronization among the slaves. This is repeated until all the subprograms at the different levels of the task graph are executed concurrently, and the final result is accumulated in the master station.; The performance of the network for this distributed program execution is characterized by the system speedup, which is defined as the ratio of the sequential execution time in a single workstation to the parallel execution in the network. The theoretical speedup equation is derived by modeling the network and considering various performance degradation factors including scheduling time, network load and size, communication time, TCP/IP communication overhead, task execution and synchronization time. The sequential and parallel execution times and the performance degradation factors were measured during the implementation for various network and station loads. The measurement values and theoretical results have shown that the system performance is degraded mostly by the heavily loaded nodes, the TCP/IP overhead (0.2 s), and the network size. |
