Optimization of KVM Network Based on CPU Affinity on Multi-cores

In order to improve the network performance of KVM virtual machines, an approach based on CPU affinity is proposed in this paper. By running the processes on the same CPU, the processes can run more efficiently by reducing performance degrading situations such as cache misses. In some performance-critical situations, it makes sense to bind the process to the same CPU. We run the benchmarks NETPERF and PING to validate our method. The results show that with the optimization of CPU affinity, the network latency of KVM decreased by 20% (taskset) and 10%(cgroup), while sustaining the same network bandwidth compared with four CPU cores. The CPU affinity method can effectively improve the network performance of KVM virtual machine by improving the cache performance and reducing the called times of some important models.