A dynamic load balancing strategy is being developed for multiprocessor system based on neural network architecture. The neural network is implemented as a central job dispatcher system, where the information on the multiprocessor architecture and processing capability of each node is known. The dispatcher assigns tasks to processors to optimize the system performance and to meet deadline dependency constraints. In addition, the job dispatcher keeps track of the overall system performance and learns the nature of the load balancing to improve the performance of the next decision. Inherent to the neural networks is their ability to learn the dynamics of the environment they are operating in through self organization. Hence, the job dispatcher is an intelligent system which learns the particular distributed computing environment and improves its performance. Anticipated benefits include real-time, optimized, fault-tolerant, dynamic load balancing for defense computing systems with intensive tasks such as: battle management, multi-target acquisition/tracking and multi-weapon control/allocation. In commercial applications, the intelligent dispatcher would result in efficient use of computing resources in banking, office automation, and academic research environment.
