Reliable and efficient hierarchical organization model for computational grid

Although the hierarchical model appears to be an effective solution to organize the resource managements in grid systems which have more stringent demand for both scalability and efficiency, it has some limitations which need to be addressed. For example, the master/manager resources in different levels represent single points of failure and they may be sources of bottleneck and communication overhead especially if they are not efficiently selected. Moreover, the dynamic and fault-prone nature of grids cannot be treated by static structures while the manual construction and repairing are also prohibitive due to the highly caused overhead which often represents a significant obstruction to an efficient resource utilization (especially for those with intermittent availabilities). The main objective of this paper is to first introduce a self-repairing n-try dynamic hierarchical grid model for scheduling and load balancing in which each master resource will be replicated on one of its children resources. Second, an efficient methodology to elect masters-replicas resources is proposed. In this methodology, the masters-replicas are selected based on both resource reliability (in terms of MTBF) and resource proximity from the other nodes in specified groups (in terms of communication latency). Validation of the proposed methodology based on the proposed model is done via simulation. Experimental results show that the proposed model has a great impact on the overall performance. Compared to other approaches, the simulations show that our approach decrements the average completion time (ACT) by 18.9%-25%, increases the tree stability ratio (TSR) up to 26.2%-27.1%, and minimizes the total communication overhead (TCO) by 4.4%-18.7% in the range of system parameter values examined.