面向多样应用和异构集群的约束调度机制  

Constraint scheduling mechanism for diverse applications and heterogeneous clusters

作　　者：王旻[1] 张章[1] 汤学海 韩冀中[1] 

机构地区：[1]中国科学院信息工程研究所信息智能处理技术研究室,北京100093 [2]中国科学院大学,北京100049

出　　处：《计算机应用研究》2015年第10期3070-3074,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(60903047);国家"863"计划资助项目(2012AA01A401);中国科学院先导专项项目(XDA06030200)

摘　　要：针对传统集群调度器在异构环境中无法处理约束的问题,提出了一种约束调度机制,包括约束描述方法和约束调度算法。该描述方法通过易拓展的属性集合方式,描述异构化的任务需求和机器属性,进而描述各种约束;该调度算法将硬约束作为过滤标准,软约束作为选择标准,综合考虑软硬约束,为任务分配最优机器。实验表明,提出的约束调度机制不仅具有良好的易用性、较短的调度响应延迟,而且可以显著提高任务执行效率。因此,该调度机制可以较好地用于异构环境的集群调度中。This paper proposed a constraint-based scheduling mechanism consisting of a constraint-description method and constraint scheduling algorithm to solve the problem of traditional cluster schedulers＇ inability to process constraint based issues in heterogeneous clusters. This constraint-description method could use extensible attribute sets to described heterogeneous task requirements and machine attributes, and then described all sorts of constraints. This algorithm used the hard constraints as filter criteria, the soft constraints as selection criteria, and then selected the best machine for task placement. Experiments show that the proposed constraint scheduling mechanism has better usability, lower scheduling response latency and significantly improved task execution efficiency. Thus, the scheduling mechanism can be better used for cluster scheduling in heterogeneous environments.

关 键 词：集群调度 约束 云计算 异构性 

分 类 号：TP393.09[自动化与计算机技术—计算机应用技术]