Load balancing in distributed multi-agent computing systems

In this paper, a dynamic load-balancing scheme is developed for implementation in an agent-based distributed system. Load balancing is achieved via agent migration from heavily loaded nodes to lightly loaded ones. The credit based concept is used for the dual objective of: (1) the selection of candidate agents for migration, and (2) the selection of destination nodes. This represents an elaboration of previous research work aimed at the selection of agents only. Multiple linear regression is used to achieve our dual objective. A complexity analysis for the proposed system operation is performed, the complete operation is O(n²) where n is the number of agents on the local host. The proposed system is implemented using JADE (Java Agent DEvelopment Framework), the multiple linear regression operation is performed using R Tool. The experimental results show a modified system operation in terms of reduced user’s query response time, by implementing agent load balancing.