Evaluation of Load Balancing Strategies
Haroon Rashid, Usman Aftab · 2004
Abstract: The distributed memory parallel processing technology encourages researchers to attack computationally intensive problems using inexpensive network of workstations. But unfortunately, distributed applications often face the problem of load imbalance in a heterogeneous environment. So, many algorithms / techniques were proposed [7 and 8] to manage / balance imbalanced load. And these algorithms differ from each other on the basis of certain controlling parameters. In this paper we have studied behaviors of a few such algorithms, i.e. fixed granularity, variable granularity (guided self scheduling) and global centralized task migration [4] algorithms. And the results yielded from this study demonstrate performance of the said algorithms, each in a certain condition. This study was carried out with matrix multiplication as benchmark application on a NOWs using PVM as parallel library.