Dynamic Load-balancing of Image Processing Applications on Heterogeneous Networks of Workstations

S. Piersall and S. Elfayoumy (USA)


Dynamic Load Balancing; Heterogeneous Networks of Workstations; Image Processing


Networks of workstations (NOWs) present a non-uniform platform, comprised of a heterogeneous mix of computing architectures. As a result of this heterogeneous quality, NOWs do not usually produce the anticipated performance due to load imbalances. Furthermore, because the execution time of applications executed on a NOW are usually bounded by the throughput of the slowest workstation, a heavily loaded workstation in a NOW, without the benefit of load-balancing, can cause significant performance degradation. This paper presents a dynamic load-balancing architecture (DYLAPSI) that focuses on image processing applications on heterogeneous NOWs. This architecture is comprised of several features that are well-suited to a heterogeneous environment, including: a flexible image partitioning scheme that assigns tasks in proportion to a workstation’s computational capabilities, an overlapping task assignment scheme that allows load balancing operations to be carried out with minimal communications cost penalties, and an image queuing mechanism that allows multiple images processing. Several experiments were developed to examine the performance gain produced by using DYLAPSI in different load scenarios. Results obtained from these experiments demonstrate that DYLAPSI can offer performance increases of up to thirty-seven percent in scenarios of heavy load imbalance.

Important Links:

Go Back