JISE


  [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16]


Journal of Information Science and Engineering, Vol. 33 No. 2, pp. 463-483


An Efficient Two-Level Hierarchy Job Scheduling and Task Dispatching Strategy for Cluster Rendering System


QIAN LI, WEI-GUO WU, ZE-YU SUN AND JIAN-HAN HUANG
Department of Computer Science and Technology
Xi’an Jiaotong University
Xi’an, 710049 China
E-mail: {qian.l; wgwu}@stu.xjtu.edu.cn; lylgszy@163.com; huangjhsx@gmail.com


    When high performance computing platforms are widely used for rendering application, scheduling of rendering jobs is crucial due to job starvation and resource fragmentation problems in the existing scheduling strategies. Meanwhile, these strategies rarely consider a tradeoff with fairness and performance. Moreover, the traditional task-cen- tered assignment method and naïve load balancing strategy cannot make full use of rendering resources. Aiming at solving these issues, an efficient two-level hierarchy job scheduling and task dispatching strategy (RF-FD) for cluster rendering system is proposed. Specifically, the reservation-based FirstFit (RF) job scheduling strategy and feedback-based tasks distribution (FD) strategy are integrated to obtain the maximization of system performance and load balancing. The RF strategy uses the ideology of reservation on low-priority jobs to maximize the resources utilization rate when the jobs are blocked. And the FD strategy uses feedback of resource usage to choose an appropriate number of threads for the renderer and then divides rendering nodes into fine-granularity rendering units to balance load distribution. With different evaluation metrics, the experimental results demonstrate that the proposed strategy outperforms the existing scheduling strategies combining with the fixed threads and naïve load balancing method while guaranteeing the fairness.


Keywords: cluster scheduling strategy, tasks assignment, rendering, resource reservation, feedback, load balancing

  Retrieve PDF document (JISE_201702_11.pdf)