Research On Flexible Multi-Path Bandwidth Scheduling Algorithms For Multiple Priority Requests In High-Performance Networks

Many applications in scientific,business,and engineering domains generate large amounts of data on a daily basis,which must be reliably transferred to collaborating sites for processing and analysis within a deadline.However,the traditional shared IP network are insufficient to provide the Quality of Service(QoS)required for such big data applications.The rapid development of Software-Defined Networking(SDN)technology facilitates the realization of bandwidth reservation in High-Performance Networks(HPNs),which have emerged as an effective solution to big data transfer.Most of the existing work for bandwidth scheduling in HPNs is focused on single-path scheduling,which has two main limitations: i)it does not scale up to meet the requirement for the rapid growth of data size;ii)it is incapable of making full use of expensive network resources.This thesis investigates the problem of scheduling a batch of prioritized and deadline-constrained bandwidth reservation requests with multiple link-disjoint paths in HPNs to minimize the average Earliest Completion Time(ECT)and Shortest Duration(SD)and maximize the user satisfaction degree and scheduling success rate,summarized as follows:(1)We propose two problems of bandwidth scheduling with flexible multi-paths for earliest completion time and shortest duration,and prove both of them to be NP-complete.(2)Following the minimum bandwidth principle and the maximum bandwidth principle,we design flexible multi-path scheduling algorithms to solve these problems.The key ideas include: i)sort bandwidth reservation requests according to the priority against data size;ii)compute multiple paths based on the minimum bandwidth principle to meet the data transfer requirement of minimum bandwidth,therefore improving user satisfaction degree and scheduling success rate;iii)utilize the largest available bandwidth based on the maximum bandwidth principle to reduce the average ECT/SD in data transfer.(3)We conduct small-and large-scale simulation experiments for performance evaluation in a real HPN topology.Experimental results show that the proposed algorithms outperform existing ones consistently in terms of user satisfaction degree,scheduling success rate,and average ECT/SD.