Design And Realization Of Web Cluster Jobs Management System

With the development of the Internet and the improvement of productiontechnology, high-performance computer cluster has covered a wide range ofapplications. With this trend, the cluster job management system has attracted moreand more people’s attention. Cluster job management system not only allows users toeasily use the cluster resources, but also improve the utilization of system resources.Torque Resource Manager maintained by the Adaptive Computing is one of currentthe most popular of high-performance computer cluster batch job management system.However, it’s difficult to use for newer, and it exist login security defects.As the background of subprojects “CYB flow river model cloud computing andservices platform” which Sun Yat-sen University cooperate with the GuangdongProvincial Department of Water Resources, this thesis presents design andImplementation of three function modules include user management, projectmanagement and operations management. The platform system adopt B/S structure toprovide users with a simple and friendly user interface, improve the utilization of thesystem and shorten the completion time of the job. In addition, the thesis proposes animproved measure of Torque.In this paper, the main tasks are as follows: First, according to the characteristicsof the platform for scientific computing, the thesis concludes the business requirementand functional requirements of the platform, and clear the system use cases. Followedby the technical assessment analysis, in order to achieve the stability and maintenanceof the platform, the thesis determines to use J2EE (Java platform Enterprise Edition)architecture and cluster job management system Torque. Furthermore, with the resultsof the requirements analysis, the platform is designed to multi-layer technologyarchitecture based on the MVC model, and is divided into four functional modulesconsist of user management, project management, job management, and operationmanagement, and illustrates the implementation of main use cases and the total classdiagram design. Then, the thesis defines platform classes and methods in details, andultimately the all platform function mentioned. Users can easily submit computing jobs, and the administrator is allowed to manage the job queue without Torquecommand, which greatly facilitates the administrator. In addition, through researchand analysis the scheduling tool Maui integrated with Torque, not only to achieve aseamless integration with the platform system, and introduce algorithm which basedon job type and amount of computation time indicator that is a prerequisite of Mauieffective job scheduling. Meanwhile, in order to solve the inherent defects of the jobestimation time in Torque, the Adaptive Job estimated time mechanism was proposed,and the effectiveness of the proposed method is confirmed by experiment.