الفهرس | Only 14 pages are availabe for public view |
Abstract Computational Grid is an attractive form of distributed systems which enables the sharing and fully exploiting of geographically-distributed, loosely-coupled, and often underutilized resources to solve the large scale applications in a faster and cheaper manner. Computational grid has different constraints and requirements compared to other distributed systems. Resource and workload managements are key grid services, where issues of allocation and load balancing represent a common concern for most grid systems. Although there are many studies have been devoted to load balancing in grids, the most of already developed load balancing approaches cannot well capture the complexity of the resources of highly dynamic grids, which are often characterized by high dynamicity, computational heterogeneity, and different reliability. This thesis focuses on developing solutions for a scalable and effective load balancing framework for highly dynamic computational grids which contain not only permanent participating resources but also intermittently available computing resources. Firstly, a self-repairing n-try dynamic hierarchical organization model was developed which enables scalable and effective resource allocation and load balancing in highly dynamic grid environments. In this model, the allocation and load balancing decision making entities and their replicas are dynamically selected using a proposed efficient methodology which takes both the reliability and the communication performance of resources into account. Secondly, a reliable and efficient MCDM-based hierarchical load balancing method using TOPSIS (a Technique for order Preference by Similarity to Ideal Solution) was presented. In this method, the allocation and load balancing are considered as a multi criteria decision making problem (MCDM) where the attributes: resource reliability, resource load index, and resource computational capability (measured by the estimated job completion time on that resource) are the criteria upon which the allocation and load balancing decisions are made. A novel weighting mechanism was also proposed in which the weights of the considered criteria are adaptively and autonomously adjusted based on the characteristics of jobs and the current state of the system. Finally, extensive simulation studies were performed to analyze the performance of both the proposed grid model and the proposed load balancing method. The simulation results revealed that they both significantly outperform pre-existing approaches that are relevant to this research. Keywords: Computational Grid, Grid Scheduling, Load balancing, Fault tolerance, Distributed System. |