Unveiling performance of NFV software dataplanes

The key technology of NFV is software dateplane, which has attracted much attention in both academia and industry recently. Yet, in practice, there is very little understanding about its performance till now. We make a comprehensive measurement study of NFV software dataplanes in terms of packet processing throughput and latency, the most fundamental performance metrics. Specifically, we compare two state-of-the-art open-source NFV dataplanes, BESS and ClickOS, using commodity 10GbE NICs under various typical workloads. Our key observations are that (1) both dataplanes have performance issues processing small (?128B) packets; (2) it is not always the best to colocate all VMs of a service chain on one server due to NUMA effect. We propose resource allocation strategies to remedy the problems, including carefully adding vNIC queues and CPU cores to vNFs, and distributing VNFs of a service chain to separate servers. To essentially address these problems and scale their performance, software dataplanes need to improve the support for NIC queues and multiple cores.