Abstract:
Task scheduling in grid computing is a challenging problem because of heterogeneous and dynamic nature of grid resources. The performance of grid resources is constantly changing. Task scheduling becomes more complicated when various quality of service (QoS) demands arise from users. QoS is an extensive concept and it varies from application to application. QoS is set of constraints for effective execution of an application.
In this dissertation work, six heuristics are proposed for independent task batch mode scheduling. Two heuristics, namely, segmented weighted time-min (SWT-Min) and segmented weighted time-max (SWT-Max) are for resource performance based independent task scheduling. Both heuristics first finds the performance metric of resources and uses this metric to find the weighted time of each task. They divide the tasks in number of segments. Each segment is assigned a priority and mapping of tasks is done in the descending order of priority of segments. Four heuristics, namely, QoS guided weighted mean time-min(QWMTM), QoS guided weighted mean time min-min max-min selective(QWMTS), priority based QoS guided weighted time min-min max-min selective(PQWMTS) and multiple QoS guided weighted mean time min-min max-min selective(MQWMTS) are proposed for QoS based independent task scheduling. The QWMTM, QWMTS and PQWMTS heuristics are for single QoS based task scheduling and network bandwidth is considered as QoS parameter. The MQWMTS heuristic is for multiple QoS based task scheduling. Response time, execution cost and priority are considered as QoS parameters for testing the heuristic. A generalized function is presented to consider all QoS and to generate a utility value of tasks. This utility value decides the order of execution of tasks.
The gridsim simulation toolkit is used to validate the proposed heuristics. The heuristics are evaluated on the basis of makespan and resource load balancing. The results show that all proposed heuristics gives good improvements in makespan and do better resource load balancing than existing heuristics such as QoS guided min-min, weighted mean time-min, weighted mean time min-min max-min selective, min-min and max-min.
