Distributed load balancing scheme with adaptive location policy based on state knowledge

Abstract

In distributed systems, the purpose of load balancing is to improve system performance by redistributing tasks from heavily-loaded nodes to lightly-loaded ones. The location policy is the most important component of a distributed load balancing algorithm. It locates the destination nodes to or from which tasks will be transferred. It should evenly distribute workload to the entire nodes with minimal delay for transferring task. An efficient adaptive location policy is required in the sense that it can react to changes in system state and achieve high performance. In this thesis, we propose a new distributed adaptive location policy based on state knowledge. The state knowledge is composed of the system state information collected at run time and the predefined global static information that is a priority order of each node for transferring tasks. The information is systematically maintained in each node by using an efficient data structure and a rule for collecting state information with low overheads. When load balancing is triggered at a heavily-loaded node, the proposed location policy dynamically predicts lightly-loaded nodes and the other heavily-loaded ones by exploiting state knowledge. Then it adaptively finds a good lightly-loaded node that minimizes useless polling and maximizes even load distribution. An analytic model is developed to compare the proposed adaptive location policy with other well known policies. The validity of the model is checked with an event-driven simulation, and it is shown that the proposed adaptive location policy exhibits a significant performance improvement over other policies, especially at high system loads. Also, the proposed policy is shown to significantly improve polling hit ratio and to avoid system instability.