An Iterative Processor Allocation Refinement Approach to Moldable Job Scheduling Guided by Resource Utilization and Job Turnaround Time

• Main Authors: YU, SHUO-TING, 余碩庭
• Other Authors: HUANG, KUO-CHAN
• Format: Others
• Language: en_US
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/4v9548

碩士 === 國立臺中教育大學 === 資訊工程學系 === 107 === With the advances of parallel programming technology, moldable jobs have become a common parallel computing model, but have not received enough research attention yet. Moreover, usage convenience and system efficiency are becoming even important in the emerging HPC as a Service (HPCaaS) model. Therefore, moldable job scheduling is a crucial issue for achieving the goal of HPCaaS regarding both user experience and service provider interests. In contrast to the straightforward processor allocation in traditional rigid job scheduling, processor allocation becomes a challenging problem in moldable job scheduling since the scheduler now has to determine the most appropriate number of processors for each job’s execution. In this thesis, we focus on exploring the issues of processor allocation in online moldable job scheduling where jobs arrive at different time instants, and propose a new iterative process allocation refinement approach guided by both resource utilization improvement and job turnaround time reduction. The new approach features three innovative methods dealing with processor allocation for simultaneous jobs, enforced processor allocation expansion, and control of iterative process, respectively. The proposed approach was evaluated with a series of simulation experiments, and compared to previous methods in the literature. The experimental results show that our approach is superior to previous methods in terms of average job turnaround time. We also developed an execution schedule visualization tool to aid our research work, which has been shown effective in the investigation of performance bottlenecks and analysis of scheduling algorithms’ behavior.