A Performance Study on Different Classes of Load Balancing Algorithms

. We present a performance study on two classes of load balancing algorithms: with and without locally maintained load state tables. We found that a load state table avoids indiscriminate pollings by exhibiting a filtering effect . However maintaining a load state table in a broadcasting channel environment does not provide performance advantage over the use of pollings. Only if channel utilization is a concern, sacrificing the performance to reduce the channel overhead may worth consideration. Keywords: Dynamic Load Balancing Algorithms, Distributed Systems, Performance Modeling 1 Introduction In a distributed system where a set of processing nodes are connected by a local area network, some nodes tend to have higher task arrival rates than others [1]. A dynamic load balancing (LB) algorithm relocates tasks among the nodes so that workload imbalance can be smoothed out, and thus the computing capacity of idle or lightly loaded nodes can be used. Many LB algorithms have been propose..