Research Article
A Calculation Method of vCPU Occupancy Rate of Virtual Machine Forwarding Process
@INPROCEEDINGS{10.4108/eai.27-8-2020.2294712, author={Hua Lu and Qinshu Chen and Jidong Zhang and Xuefei Duan}, title={A Calculation Method of vCPU Occupancy Rate of Virtual Machine Forwarding Process}, proceedings={Proceedings of the 13th EAI International Conference on Mobile Multimedia Communications, Mobimedia 2020, 27-28 August 2020, Cyberspace}, publisher={EAI}, proceedings_a={MOBIMEDIA}, year={2020}, month={11}, keywords={dpdk sr-iov ovs-dpdk nfv vcpu vm occupancy rate forwarding process}, doi={10.4108/eai.27-8-2020.2294712} }
- Hua Lu
Qinshu Chen
Jidong Zhang
Xuefei Duan
Year: 2020
A Calculation Method of vCPU Occupancy Rate of Virtual Machine Forwarding Process
MOBIMEDIA
EAI
DOI: 10.4108/eai.27-8-2020.2294712
Abstract
Network Function Virtualization (NFV) is a technology that implements network functions through virtualization on x86 universal servers. In order to improve the forwarding performance of a virtual machine, usually the most effective method is to use multiple cores and exclusive methods of the vCPUs in the virtual machine to monopolize the physical CPU resources for multiple forwarding processes to use independently. The forwarding process discards the interrupt-based asynchronous signal sending mechanism to avoid the impact of interrupt switching on the forwarding overhead. Instead, it uses a while 1 dead-loop to poll the packet receiving queue. Once there is a packet in the packet receiving queue, the packet is immediately forwarded. Because the vCPU bound to the forwarding process works in an infinite loop polling mode, the vCPU occupancy display is 100% regardless of whether it is in the no-load or full-load phase. Because the VM cannot obtain the real load of each forwarding vCPU in real time and expand it in time, it will cause VM to lose a lot of packets due to overload operation, which will affect the quality of the service carried by the VM. This article studies how to measure the real utilization of vCPU in real time. The results show that the real vCPU occupancy can be accurately calculated using this solution.