摘要
随着虚拟网络的规模变大和虚拟网络的功能增多,单台服务器的资源可能无法支持一个较大规模NFV网络的运行,需要将其划分为多块并分别部署到服务器集群上,而且要使该NFV网络呈现出良好的性能.本文通过网络测量手段研究了NFV网络的资源需求和宿主服务器的资源供给以及两者的映射关系,探讨了利用多权值约束的图划分算法描述NFV网络中的关键计算资源的问题;提出了一种基于Metis划分NFV网络算法(Partitioning NFV Networks based on Metis,PNFVNbM),该算法在划分前对多权值进行融合,在划分后评估划分效果以改进融合参数,以提高各个部分的平衡度.最后建立了原型系统,对多个大规模NFV网络进行了实验验证.实验的结果表明了PNFVNbM算法能够均衡地将大规模NFV网络划分在多台服务器上,使得服务器集群能够支撑大规模NFV网络及其应用.
The widespread application of NFV technology in cloud computing and data center networks has made the scale of virtual networks larger and the functions of virtual networks increased.According to results in a production environment,the resources of a single server may not be able to support the operation of a large-scale NFV network.A key technical challenge is to divide a large scale NFV network into several parts and deploy them to a server cluster that each server runs a subnet of the NFV network.The multi-level partition algorithm Metis proposed by Karypis and Kumar is considered to be a very effective weighted graph partition algorithm,which can achieve the balance of the number of edge cuts and each partition.However,there are still many specific technical problems that need to be solved in the scheme for applying the Metis algorithm to the division of NFV networks,such as reasonably representing resources as weights in the diagram,and how to reflect the characteristics of NFV networks.There have been some researches on the algorithm of Metis to divide the network topology at home and abroad,but there is still a lack of research on the use of graph division algorithms to deploy large-scale NFV networks.Based on this idea,this paper studies the approximate resource requirements of servers and NFV networks through network measurement methods and discusses the problem of describing key computing resources when using the Metis graph partitioning algorithm.We quantified the CPU,memory,and disk usage of typical applications of NFV networks built using LXC and analyzed the CPU usage of the forwarding chain consisting of multi-hop virtual routers.According to the measurement results,we believe that the CPU of the host server is a critical shortage of resources in the system.Based on some modeling analysis,an NFV network partitioning algorithm based on a multi-weight constraint called Partitioning NFV Networks based on Metis(PNFVNbM for short)is proposed.This algorithm fuses multiple weights before partitioning and evaluates them after partitioning.Divide the effect to improve the fusion parameters to improve the balance of each part.Finally,we used a popular virtualization technology,LXC,to build a fairly large-scale NFV prototype system,and deployed multiple NFV networks on the system to test our partitioning algorithm.We compared the performance of NFV networks with regular and irregular shapes and different numbers of subnets.In addition,we also measured connectivity and communication bandwidth between parts deployed on different servers.The experimental results show that the PNFVNbM algorithm can evenly partition a large scale of NFV network.Large-scale networks divided by this method can ensure the performance of the network and make the work load-balanced on all available servers,which mean that this method makes the server cluster support the large-scale NFV network and its applications effectively.In the future,we will use this method to study the technology of deploying large-scale NFV networks on server clusters automatically and parallelly,and do some research on processing technologies when the workload changes dynamically.
作者
邓理
戴宁赟
许博
邢长友
陈鸣
DENG Li;DAI Ning-Yun;XU Bo;XING Chang-You;CHEN Ming(College of Computer Science and Technology,Nanjing University of Aeronautics and Astronautics,Nanjing 211106;College of Command Control Engineering,Army Engineering University,Nanjing 210007)
出处
《计算机学报》
EI
CSCD
北大核心
2020年第10期1958-1968,共11页
Chinese Journal of Computers
基金
国家自然科学基金项目(61772271,61379149)资助.
关键词
NFV网络
Metis算法
资源描述
多权值约束
服务器集群
Network Function Virtualization network
Metis algorithm
resource description
multi-weight constraint
server cluster