Latency-aware interrupt rate control for network I/O virtualization

摘要

Virtualization technologies are playing an increasingly important role in today's new data centers. The virtualized network resources are becoming the main factor in cloud computing environments, and many researches take efforts to increase the network I/O virtualization performance. Single Root I/O Virtualization (SR-IOV) is a popular technology inherited from direct I/O, and it's able to approve nearly bare metal performance. However, previous researches on SR-IOV all focused on improving throughput and reducing CPU utilization, but had no work on the latency limitation requirement of different applications. In this paper, we analyze the relationship among one VM's network throughput, latency and CPU utilization by several experiments in SR-IOV environment. Based on our analysis, we propose the design of a latency-aware interrupt rate control method to provide high network throughput while ensuring the limited latency of different applications.