A Job Management Framework Enabling Deployment of Allocation Policy Regarding Computational and Network Resources

Yasuhiro WATASHIBA  Yoshiyuki KIDO  Susumu DATE  Hirotake ABE  Kohei ICHIKAWA  Hiroaki YAMANAKA  Eiji KAWAI  Haruo TAKEMURA  

D - Abstracts of IEICE TRANSACTIONS on Information and Systems (Japanese Edition)   Vol.J97-D   No.6   pp.1082-1093
Publication Date: 2014/06/01
Online ISSN: 1881-0225
Type of Manuscript: Special Section PAPER (Special Section on Internet Technology and its Applications)
job management system,  Software Defined Networking,  OpenFlow,  

Full Text(in Japanese): PDF(1.8MB)
>>Buy this Article

High-performance computing environment exemplified by cluster systems shares its resources among multiple user jobs. To reduce execution time of parallel and distributed computation on such an environment, efficient allocation of computational and network resources to each job is necessary. Most of job management systems, or JMSs available today, however, determines resource allocation by relying on only status information on computational resources. In this paper, we propose a novel job management framework that enables resource allocation of both computational and network resources based on user-specified resource requests for jobs. Technical feature of the proposal is that the proposed framework is built on Open Grid Scheduler/Grid Engine (OGS/GE), an existent JMS with OpenFlow, which allows dynamic control of a network in a central way. We confirmed that execution time of jobs was reduced using the proposed JMS framework in both cases of bandwidth and latency preferred resource allocation policy.