Cooperative Scaling Of Service Function Chain In Network Function Virtualization

As the fluctuation of human group behavior in space and time,the traffic and the number of users carried by network services rise and fall.In data center network and mobile core network,network services are generally provided by the service function chain(SFC)that is composed of multiple network functions(NFs),such as firewall,intrusion detection system(IDS),mobility management entity(MME),serving gateway(SGW),etc.The emerging technologies,such as network function virtualization(NFV)and software defined network(SDN),make it possible that the SFC could scale out and in elastically and dynamically,meeting the service requirements,avoiding the waste of resources and reducing capital expenditures(CapEx)and operating expenditures(OpEx).But there are several problems need to be solved.In order to avoid the violation of service level agreements(SLAs)and guarantee the consistencies of NF internal states,when multiple instance pools of NFs in the SFC need to be scaled out,part of the data flows and their corresponding NF internal states need to be migrated to new instances.As during the migration of a data flow,the new arrived packets of this flow will be forwarded to the network controller for buffering.Then after the migration completes,the buffered packets will be forwarded to the new instance in order.So during this process,the buffered packets will employ the memory of the controller and a little latency will be introduced to the migrated flow.What's worse,the additive effect caused by the scaling of multiple NF instance pools will multiples the introduced latency,and the memory of the controller will be occupied by large amounts of buffered packets,which increases the potential risk of packet loss caused by buffer overflow.So the challenge,how to eliminate the additive effect,needs to be addressed.This paper focuses on the flow selection problem encountered before the flow migration during the scaling and load balancing of the service function chain.When multiple instance pools of NFs are required to be scaled out and the loads among instances of every scaled NFs need to be balanced to some extent as soon as possible,an amount of data flows traversing these NFs need to be selected to migrate.Then the problem is selecting which subset of the existing flows to migrate to new instances,making the total number of migrated flows the least and minimizing the total buffered packets and introduced latency.The main contributions of this paper are as follows.(a)This paper describes the flow selection problem encountered before the flow migration during the scaling of multiple NFs in the SFC and models it as an integer non-linear optimization problem from the perspective of cooperative scaling with a global com-prehensive consideration.(b)Due to the complexity of this problem,it's difficult to obtain the optimal solution in a limited time.So we propose a priority-based preference selection algorithm to find the feasible optimized solution quickly.(c)The heuristic algorithm is evaluated by real traffic trace from data center network with several NFs.The results show that the algorithm proposed in this paper can signifi-cantly reduce the total number and packet rate of flows need to be migrated than scaling one by one and random selection algorithm,decreasing the total number of packets af-fected by buffering and minimizing the total latency introduced by flow migration.