No abstract

Jobs scheduling and load balancing functionalities are crucial for best Grid performance and utilization. In this paper, we present a dynamic and adaptive polyhedron topology-aware Grid load balancing method which we called TunGrid. Its main objective is the extreme decentralization of the scheduling and the load balancing processes at the lowest processing and networking cost. It ensures load balancing through an adaptive local neighborhood propagation strategy of overload. TunGrid was experimented using the GridSim simulator and compared to other methods. Experimental results are presented and discussed.

Today's computing development is being characterized by the rapid development of high speed networks and the increase in computing power. Computing is not any more limited to the supercomputers, PCs and laptops but also smart phones and tablets which are available for billions of users offering high computing performances at low cost and interconnected via Internet. This continuing technological development is leading the increase importance of the distributed computing paradigms and the apparition of new ones. This paper aims to review the most important distributed computing paradigms and the principal similarities and differences between them. This survey is a kind of a brief road map that would be useful for researchers, students, and commercial users.

No abstract

KEYWORDSGrid computing, meta-scheduling, distributed load balancing, topology-aware. 1.INTRODUCTIONGrid computing[1] extends existing distributed computing to a more unified and collaborative structure. It enables multiple computing resources, that are geographically distributed and owned by different individuals or organizations, to be logically coupled to form the illusion of a powerful super computer with infinite capacity [2].Scheduling and load balancingare essential grid software infrastructure services. Theseservices assigntasks to nodes(i.e. computing resources) and transfer tasks from overloaded to under-loaded nodesin orderto maximizeresource utilization on one hand,and minimizing the total task's execution time on the other hand. Considering the grid load balancing problem, several solutions have been proposed using different methods and strategies, e.g., [3] Among the most critical issues pertaining to grids is how to manage the resources [4]. In the proposed solution, the grid resources are organized according to a logical topology called theresource managementmodel [4]. This topology/model defines how the different grid entities communicate and work in order to achieve the received jobs/tasks. R. Buyyaet al. proclaimed that the choice of the right model for the resource management architecture plays a major role in its eventual (commercial) success [4].The motivation behind this paperis based on the observation that the load balancing problem can be improved by simply improving the classical used models tree, star, and P2Pand further proposing a new one; a polyhedron sphere like model that was introduced for the first time in [14]. This logical model has the merit of providing optimized load balancing in a stateless context. It guarantees a high level of decentralization and scalability of the load balancing process leading tobetter resource utilization rates and minimum tasks execution time. Therefore,this work proposes adecentralizedtopology-awareload balancing solutionfor independent tasks in a grid environment