A Study Of Multiple Applications Scheduling Algorithms In Optical Network Based Distributed Computing System

There are more and more advanced applications such as e-science applications and collaborative designs, so we need a computing infrastructure for these advanced applications. Geographically distributed resources can be integrated by network in dynamic, distributed virtual organizations to solve complex problems. On the other hand, optical networking can offer huge capacity and relatively low latency, as well as dynamic control and allocation of bandwidth at various granularities.In this paper, we make a study on multiple applications scheduling algorithms and protection policies in optical network based distributed computing system. In the optical network based distributed computing system, we connect computing resources with optical network. The essential aspect for optical network based distributed computing system is scheduling, which is the special and temporal assignment of the tasks of applications to the processors of the target system. By scheduling, multiple applications from different users can be executed on one optical network based distributed computing system simultaneously. On the other hand, when there are fails in optical links, we should propose protection policies for the system.Based on the total execution time and fairness, we propose three scheduling algorithms about multiple applications. In Multiple Direct Combination algorithm, we schedule all tasks of applications in decreasing order of their bottom levels. In Multiple Delay-Weighted Combination algorithm, we add an index to the bottom levels of former applications'tasks in order to change the scheduling order of tasks. In Multiple Insert algorithm, we use free intervals of the system to schedule a new application. According to simulation results, we can see these three algorithms make tradeoff in the total execution time and fairness.We propose Multiple Deadline algorithm to schedule multiple applications, and each application has a deadline. During the scheduling, a new application should be scheduled as soon as possible with less impact for former applications. According to simulation results, we can see Multiple Deadline algorithm has better performance.In the last, we make a study on protection policies of the scheduling algorithm about multiple applications, and propose two protection policies. During the scheduling, the multiple applications scheduling algorithm adds a protection path to the data transport in order to protect the system against optical link fails. According to simulation results, we can see the multiple applications scheduling algorithm with a protection policy has better performance when there are optical link fails in the system.