An Adaptive Load Balancing Algorithm for Systems with Bursty Task Arrivals

Usually heuristic-based load balancing algorithms cannot provide satisfactory performance with bursty task arrivals because they assume stable arrival patterns. In this paper, we present an adaptive load balancing algorithm, which employs the new Batch Transfer Approach. This approach allows a number of tasks to be transferred as a single batch, coupled with a protocol to obtain mutually agreed batch size between a sender and a receiver. Simulations show that: (1) In terms of the system mean task response time, our algorithm provides significant improvement when the system is not saturated. (2) Our algorithm can always improve the system predictability. (3) Our algorithm ensures a stable range of both mean queue length and mean task response time. Keywords: Dynamic Load Balancing Algorithms, Bursty Task Arrivals, Performance Modeling 1 Introduction In a distributed computing system where a set of processing nodes are connected by a local area network, some nodes tend to have higher..