DEVELOPMENT AND PERFORMANCE EVALUATION OF A LAN-BASED EDGE-DETECTION TOOL

This paper presents a description and performance evaluation of an efficient and reliable edge-detection tool that utilize the growing computational power of local area networks (LANs). It is therefore referred to as LAN-based edge detection (LANED) tool. The processor-farm methodology is used in porting the sequential edge-detection calculations to run efficiently on the LAN. In this methodology, each computer on the LAN executes the same program independently from other computers, each operating on different part of the total data. It requires no data communication other than that involves in forwarding input data/results between the LAN computers. LANED uses the Java parallel virtual machine (JPVM) data communication library to exchange data between computers. For equivalent calculations, the computation times on a single computer and a LAN of various number of computers, are estimated, and the resulting speedup and parallelization efficiency, are computed. The estimated results demonstrated that parallelization efficiencies achieved vary between 87% to 60% when the number of computers on the LAN varies between 2 to 5 computers connected through 10/100 Mbps Ethernet switch.