A User-Level NUMA-Aware Scheduler for Optimizing Virtual Machine Performance

详细信息    查看全文

• 作者：Yuxia Cheng (18)
  Wenzhi Chen (18)
  Xiao Chen (18)
  Bin Xu (18)
  Shaoyu Zhang (18)
  
• 关键词：NUMA ; virtual machine ; scheduling ; memory locality
• 刊名：Lecture Notes in Computer Science
• 出版年：2013
• 出版时间：2013
• 年：2013
• 卷：8299
• 期：1
• 页码：47-61
• 全文大小：1501KB
• 作者单位：Yuxia Cheng (18)
  Wenzhi Chen (18)
  Xiao Chen (18)
  Bin Xu (18)
  Shaoyu Zhang (18)
  
  18. College of Computer Science and Technology, Zhejiang University, Hangzhou, China
  
• ISSN：1611-3349

文摘

Commodity servers deployed in the data centers are now typically using the Non-Uniform Memory Access (NUMA) architecture. The NUMA multicore servers provide scalable system performance and cost-effective property. However, virtual machines (VMs) running on NUMA systems will access remote memory and contend for shared on-chip resources, which will decrease the overall performance of VMs and reduce the efficiency, fairness, and QoS that a virtualized system is capable to provide. In this paper, we propose a “Best NUMA Node” based virtual machine scheduling algorithm and implement it in a user-level scheduler that can periodically adjust the placement of VMs running on NUMA systems. Experimental results show that our NUMA-aware virtual machine scheduling algorithm is able to improve VM performance by up to 23.4% compared with the default CFS (Completely Fair Scheduler) scheduler used in KVM. Moreover, the algorithm achieves more stable virtual machine performance.