Power-saving Network Function Virtualization On Multi-processors

Network Function Virtualization(NFV)separates network functions from dedicated hardware devices to decouple software and hardware,enabling network administrators to implement networks features and its dynamic and flexible deployment based on common computing,storage,and network devices.Since NFV can dynamically allocate physical device resources,with the fluctuation of traffic processing requirements,network operators can reduce their energy consumption by reducing the number of physical devices running in a part of the time.However,in order to ensure NFV processing performance,administrators often need to keep physical devices running in high-performance mode,which will waste a lot of power when traffic processing requirements are low.In order to solve the above problems,this work designs a NFV deployment mode with centralized control based on multi-core processor,which can reduce the NFV energy consumption when the traffic processing demand is low,and improve the NFV processing performance when the traffic processing demand increases.This deployment can reduce the energy consumption of NFV on multi-core general processors when there is less demand for traffic processing.In the system designed in this work,the controller can monitor and evaluate the system traffic and load status in real time,and use the process combination mechanism and the separation mechanism to schedule the work process,so that the work process is concentrated on fewer processor cores to save energy.A preliminary evaluation using the energy analysis software powerstat shows that the NFV packet processing system using this deployment mode can reduce CPU power consumption by approximately 18%during low traffic load hours on x86 servers with 40 CPUs(Central Processing Unit)physical cores.This work summarizes the research background and current status of NFV energy conservation,and introduces the related concepts of general processor energy saving and network traffic periodicity.After quantifying the performance of energy management by processor DVFS and CPU multi-core scheduling,and the performance of extended KL algorithm,this work designs and implements NFV centralized control system.At last,this work tests and evaluates the energy consumption and performance of the NFV power saving scheduling system.The main innovations and innovative achievements of this work are:(1)In this work,the IP packet processing scenario using multi-core general-purpose processor is used to quantify the energy management capabilities of processor DVFS and CPU multi-core scheduling under various workloads,which provides the data support for NFV's energy-optimized deployment on general-purpose processors.(2)The NFV deployment method designed in this work adopts a centralized control architecture,which uses the traffic distribution information and input queue length to implement traffic monitoring and evaluation for each VNF module,and uses DVFS technology and processor affinity mechanism for physical device resources scheduling,which enables optimal deployment of multiple VNF processing chains on multi-core general-purpose processors.(3)In this work,a KL algorithm(Kernighan/Lin Algorithm)'s variant,extended Kernighan/Lin algorithm is used in the work process separation mechanism to improve system processing performance.