Load Balancing in Distributed Computing Systems Using Fuzzy Expert Systems Author Dept. Comput. Eng., Alexandria Inst. of Technol. Content Type Conferences On Page 141 - 144 Appears Modern Problems of Radio Engineering, Telecommunications and Computer Science, 2002. Proceedings of the International Conference Date 07 八月 2002 Speaker Jyue-Li Lu Abstract In this paper we present fuzzy functions to model load balancing in distributed computing systems. Due to lack of communication delays in message passing, a complete and consistent view of the entire system may never be available to a node of the system. Most of the previous work in load balancing and distributed decision making in general, do not effectively take into account the uncertainty and inconsistency in state information. Abstract Also both the load patterns and task lengths descriptions are based on crisp values which may be inconvenient. Most of the research that used fuzzy descriptions considered fixed fuzzy function parameter values that only can match in fixed load patterns. Abstract The aim of this paper is to solve the uncertainty problem in state information and task selection for migration using proposed fuzzy functions of variable membership values, and to propose an expert system that uses these fuzzy functions in decision making. INTRODUCTION Distributed computing systems have become a natural setting in many environments for business and academia. In a typical distributed system setting, tasks arrive at the different nodes in a random fashion. This causes a situation of non-uniform loading across the system nodes to occur. INTRODUCTION Loading imbalance is observed by the existence of nodes that are highly loaded while others are lightly loaded or even idle. Dynamic load balancing involves the reallocation of tasks to processors after their initial assignments. This is done by migrating tasks from the overloaded nodes to other lightly loaded nodes to improve the overall system performance. Such an assignment problem has been shown to be NP-complete. System Model Host Model 、 A communication costs table Task Model A load Table Tasks are assumed to be independent. The proposed approach requires the existence of a compile and run-time libraries support for the estimation of execution times, remaining times, penalties, communication costs, and the unfinished work. Assumptions light load, moderate load or heavy load. Load Description Load Description Previous work in dynamic load balancing using fuzzy expert systems used constant fuzzy set parameters. This might be inefficient because any load, that is described to be high with a membership value of 1, may not guaranteed to be the highest load over the network all the time. So the parameter describes the highest load must be adaptive according to the load patterns over the network. Load Description figure 1, the region from a0 to p is divided into equal six regions. Each region with length (p - a0)/6. TASKL ENGTHD ESCRIPTION The task length can be expressed as a member of one of five fuzzy sets, very long length tasks set, long length tasks set, intermediate length tasks set, short length tasks set and very short length tasks set. WORKING MEMORY INPUT DECISIONS INFERENCE ENGINE KNOWLEDGE BASE KNOWLEDGBEA S Rule[l] IF the node is lightly loaded THEN the node is a receiver. Rule[2] IF the node is moderately loaded AND the number of heavily loaded nodes is nearly equal to the number of lightly loaded nodes THEN the node is neutral. Rule[3] IF the node is moderately loaded AND the number of heavily loaded nodes is more than the number of lightly loaded nodes THEN the node is receiver. Rule[4] IF the node is moderately loaded AND the number of heavily loaded nodes is less than the number of lightly loaded nodes THEN the node is sender. KNOWLEDGBEA S Rule[5] IF the node is sender THEN select a receiver as a migration partner. Rule[6] IF the node fails to find a migration partner THEN the node is neutral. Rule[7] IF the node is a sender THEN select a suitable task to transfer. Rule[8] IF the node fails to select a suitable task to transfer THEN select another migration partner. CONCLUSION The proposed approach has been studied by means of simulation. Performance measure in terms of the mean response time is improved in all cases with different percentages. The application of the proposed approach on hard real time systems, using neural networks will be considered in future work. End