Load balancing algorithm based on aspect oriented approach for a multi agent distributed system

Abstract

In this paper, we present a new load balancing system based on aspect oriented approach for multi agents distributed systems. The proposed model is based on using a dispatcher agent assigned to dynamically distribute tasks received in random requests coming from agent producers. We suppose that tasks are assigned to be executed by workers agents deployed over the nodes of the distributed system. To make the initial mechanism of task distribution balanced, we propose a separated aspect where we propose a load balancing procedure assigned to be affected to the dispatcher by an aspect weaving in compilation or run times. The load balancing aspect increases the performances of the dispatcher agent using code advices in different join points. The task distribution code advice is used to implement the load balancing algorithm which determines the under loaded worker agent who will be assigned to execute the current task Ti. In this algorithm, we take into account the heterogeneity of the processing performances, the communication latency, occupancy rate of each node and the complexity of the current task. In this model we can also use metadata of the tasks to estimate more accurately the required times to perform each task. This mechanism allows predicting the performance offered by the distributed system according to the selected computational model. To describe the proposed approach, some communication diagrams, the established mathematical model and an application example are presented to prove the effectiveness of the proposed model. 146 Mohamed Youssfi et al.