基于LT-backfilling算法的集群作业调度系统

Cluster Job-scheduling System Based on LT-backfilling Algorithm

在胖节点集群系统中,目前的reservations,backfilling等主流调度算法未能充分考虑单节点过载情况。该文在传统backfilling算法基础上,充分考虑节点当前负载,将预约、回填策略相结合,提出了一种新的LT-backfilling调度算法。设计了基于XML业务流程模板的二层作业调度系统,用户不仅可以提交单个作业,而且可以提交一组具有数据相关性的作业流,使得作业调度系统更加实用化。

In a SMP cluster job-scheduling system, most of the well-rounded algorithms such as reservation, backfilling, don＇t pay much attention to the overload of system. This paper presents a LT-backfilling algorithm combined with reservation and backfilling. This algorithm focuses on the balance of system＇s load and improves the efficiency of balance of system＇s load. Based on the algorithm, this paper proposes a job-scheduling system based on XML flow telmplate, in this system, user not only can submit single job, but also can submit a data-relative job flow, the function makes this job-scheduling system more practical.