الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 130 |  |  |
| | | from | 130 | | |
Abstract
Grid systems – is a unified collection of resources connected by a network, have potential option to deliver high performance for many applications, and many system users. Achieving high performance in a grid system requires effective resource scheduling. Grid task scheduling is one of the most important parts in Grid resource management system. This thesis introduces a proposed task scheduling algorithm which is proposed and implemented based on Min-Min algorithm with considering of load balancing. This proposed algorithm works by executing the small tasks by slower resources while executing the relatively large tasks by faster resources under the decided makespan. Makespan and CPU utilization are the two metrics which used to evaluate the performance of the proposed algorithm rather than improvement ratio and usages rate of resources. The proposed algorithm consumes the same running time of original Min-Min algorithm even it used a new derived Expected Sum Completed Time (ESCT) metric rather than using the standard Expected Completion Time (ECT) and Expected Execution Time(EET). The goals of this thesis are firstly reducing makespan, secondaly improving resource utilization, thirdly balancing the overload of the resources. In the first and second goal,the proposed algorithm reduces the makespan and optimizing resource utilization in the most case with approximation (90%) and (95%) respectively in dataset 1, Improved average resource utilization at all cases reached (100%) of evaluation and scored (85%) improvement rate over makespan in data set 2, the proposed algorithm scores (100%) of makespan improvements and average resource utilization overall evaluation cases, using dataset, which was found from master thesis. Different submitted tasks with different instruction processing and different data. Also, different available resources with different processing speed capabilities, and data transfer rate are presented. The proposed algorithm produces a minimum makespan on different scheduling applications compared with other algorithms. Experimentations results show that the proposed algorithm produces relevant equivalent heavy resources as the guaranteed load balance schedule, and makespan is reduced.