Scientific computing using pPVM

Execution time of an application running on a cluster of workstations is critically dependent upon the performance of the underlying communication network, especially when the application demands for large data transfers among workstations. In this research, we use parallel network approach to improve the communication network performance and demonstrate this in a real cluster computing environment using multiple Ethernets. We modified the PVM environment to pPVM to schedule the application data onto multiple networks, the scheduling strategy being either round-robin or adaptive. The adaptive strategy performs well for both equal and unequal distribution of background load on the channels, and also for time-varying background load. © 1994 IEEE.