基于时间序列的预拷贝算法的改进

龙源期刊网 http://www.qikan.com.cn

基于时间序列的预拷贝算法的改进

作者：石璐 张艳

来源：《商情》2015年第46期

【摘要】虚拟机在线自动迁移需要对云数据中心各物理节点及其上的虚拟机进行实时负载监控。当某个节点出现持续负载过重时，就应触发虚拟机的迁移动作，以平衡负载，提高节点响应性能。我们一方面对当前的资源使用监控系统进行了改进，提出了一种新的专门针对虚拟机内存使用情况的检测方法。另一方面，我们引入了Hadoop框架对大规模数据中心产生的大量资源使用监控数据进行处理。由于Hadoop非常适合于分布式存储和计算，这大大提高了我们进行节点负载热点的检测速度，并且具有很强的可扩展性，克服了传统方式的处理缺陷和瓶颈。

【关键词】云计算，虚拟化环境，负载热点检测，虚拟机迁移

【Abstract】to trigger the automatic live migration of virtual machine， we need to periodically monitor the resource usage of physical/virtual nodes deployed on cloud data center. Once one node is in heavy workload for a certain duration time， we should trigger the migration task， to balance the workload and improve the response performance. However， current researches do not pay much attention on resource monitoring and workload of massive data center. Also the task of dealing with monitored data is simply conducted on single-node. Thus， this essentially does not meet with the need of large data center

we improve current resource monitoring system by proposing a novel memory monitoring

approach for virtual machine， and we only focus on the memory monitoring. It is already sufficient to monitor other resource usage using existing tools. Additionally， we introduce Hadoop framework and MapReduce parallel programming model to store and deal with the large volume of monitoring resource data. Taking the advantage of distributed storage and computing of Hadoop， we can accelerate the workload hotspot process， and furthermore we scale workload hotspot detector well for large data center which would generate large volume of monitoring data. This method also overcomes the bottleneck of single-node based workload hotspot detector.

【Keywords】cloud computing， virtualized environment， workload hotspot detection， virtual machine live migration 1相关工作

关于在线虚拟机迁移的问题，有许多学者对此进行了研究。Timothy Wood等提出了一个叫做Sandpiper的原型系统，主要聚焦于虚拟机迁移的黑盒策略和灰盒策略。更具体的来说，Sandpiper首先用黑盒策略和灰盒策略来监测位于物理节点上的每台虚拟机的资源使用情况，然后将收集到的监测信息发生给热点检测程序。该热点检测程序会分析每个物理节点和虚拟机

本文来源：https://www.bwwdw.com/article/ojat.html