Towards power-aware network function virtualization on multi-core processors

Network Function Virtualization (NFV) aims to improve network agility and reduce deployment costs by using commercial off-the-shelf hardware, typically Intel x86-based servers. However, compared with ASIC or NP-based packet forwarding engines, general-purpose processors generally provide much lower performance per watt. In this work, to improve the energy efficiency of NFV, we propose VNFMotion, a power-aware CPU allocator, by exploiting the power proportionality of x86-based multi-core processors. Specifically, the allocator can aggregate the workloads onto fewer CPU cores and put the resulting idle CPU cores into sleep for power conservation. On receiving CPU overload notification, the allocator runs a VNF graph partitioning algorithm and migrates the workloads to the newly awakened CPU core. Preliminary evaluation using Linux powerstat shows that on a 40-core x86 server, VNFMotion can reduce the CPU power usage by 19.1% during light traffic.

[1]  S.,et al.  An Efficient Heuristic Procedure for Partitioning Graphs , 2022 .

[2]  Michael L. Scott,et al.  Energy-efficient processor design using multiple clock domains with dynamic voltage and frequency scaling , 2002, Proceedings Eighth International Symposium on High Performance Computer Architecture.

[3]  Bin Liu,et al.  GreenTE: Power-aware traffic engineering , 2010, The 18th IEEE International Conference on Network Protocols.

[4]  Yang Li,et al.  Towards Zero-Time Wakeup of Line Cards in Power-Aware Routers , 2014, IEEE/ACM Transactions on Networking.