Priority Based Resource Management Techniques for Distributed Systems.

摘要

This thesis focuses on priority based resource management. Matchmaking, involving the allocation of requests to resources is a key problem in the context of resource management on distributed systems including grids and clouds. To provide quality of service (QoS) guarantees to different clients, it is crucial to support Advance Reservation (AR) requests. Most existing works do not associate priorities with ARs. Sometimes these dynamic systems also require emergency response that demand immediate action. This thesis focuses on matchmaking strategies for handling AR requests with multiple priorities, as well as Emergency requests with the purpose of achieving high system performance. Accommodating a high priority AR request can lead to the abortion of low priority AR requests, whereas the arrival of an Emergency request may require the pre-emption of an executing request. This research includes the investigation of four abortion and four pre-emption policies. A performance analysis based on simulations is conducted. A comprehensive analysis of the results leading to key insights into system behaviour and performance is presented.