Parallel processing in a local area network

Abstract

Parallel computers can take many different architectural forms. One cost effective and convenient means to acquire parallel processing is via a network of computers. Today's high-speed Local Area Network (LAN) interconnecting islands of general-purpose workstations results in vast computational power potentially available on the network. The utilisation of these networked workstations as an inexpensive parallel processing platform is realistic. There is a need to understand the behaviour of parallel processing in a LAN environment, especially its performance under different LAN traffic patterns. Of particular interest is how frame sizes, inter-packet spacing times and network utilisations impact the speedup of various problem sizes in parallel processing. This dissertation attempts to discuss all the above issues and present the results obtained under different simulated network traffic patterns.