摘要： 考虑到成本和灵活性等因素,利用公共云计算系统的资源构建临时私有云代替本地物理集群,已成为用户在特定时间段内处理大量并行作业的一种有效途径。然而现有作业调度算法不能感知并利用临时私有云的逻辑拓扑结构,容易降低紧耦合并行应用的性能。为解决该问题,提出一种逻辑拓扑感知的作业调度算法,根据私有云中虚拟机之间的带宽和延迟等信息构建逻辑拓扑关系,并综合考虑用户所提交作业的类型,进而产生对应的作业调度决策。实验结果表明,该算法在提高临时私有云中紧耦合并行应用性能方面优于开源云计算系统中常用的轮转算法和随机算法。
关键词:
临时私有云,
并行应用,
虚拟机,
拓扑结构,
调度算法,
网络性能
Abstract: Considering the cost and flexibility,building a temporary private cloud in public cloud to replace the local
physical cluster is an effective way for users to run a large number of parallel jobs during a particular period. However,the existing scheduling algorithms are not suitable for temporary private cloud hosting tightly coupled parallel applications, because they can not use the logic topology information of temporary private cloud,which probably results in the performance degradation of such type of applications. To mitigate such impact, a logic-topology-aware scheduling algorithm is presented. It builds the logic topology of virtual machines according to bandwidth and latency among them, makes scheduling decisions based on the logic topology and the information of applications. Experimental results show that compared with Round Robin and Random algorithms of open-source cloud software,logic-topology-aware scheduling algorithm improves the performance of tightly coupled parallel applications.
Key words:
temporary private cloud,
parallel application,
virtual machine,
topology structure,
scheduling algorithm,
network performance
中图分类号: