Research On Switch Architecture Based On Service Function Chain

With the continuous expansion of network scale,the network applications are becoming increasingly diverse.The problems such as poor function expansibility and resource over-allocation of traditional switches need to be resolved by fundamentally redesigning the switch architecture.Network Function Virsualization(NFV)enables data traffic to be processed through Service Function Chian(SFC)on demand by virtualizing and integrating traditional private network equipment.By introducing NFV and SFC into the switch architecture,the flexible organization and on-demand control of multiple network functions within the switch node can be realized.Thus,this thesis studies the design of the switch architecture after introducing NFV and SFC,and researches the problems of network function deployment and packet scheduling in this architecture,which can realize a more intelligent data plane and ensure the performance at the same time.This thesis proposes a Service Function Chain based Switch Architecture(SFCSA).By separating the specified functions from the traditional line card,this architecture provides flexible,customized and resource-on-demand processing functions through the service function chain in the network function pool for the switch data plane.The classifier divides the packets entering the SFCSA into different types by rule matching and encapsulates the packets that need to be processed through the network function pool using the Switch Function Header(SFH)that identifies the service function paths.The scheduler performs one or more scheduling on all packets to realize network function processing and switch forwarding in SFCSA.Aming at the problem of network function deployment in SFCSA,this thesis uses the optimization theory to model it into Binary Integer Programming(BIP),and proposes a new deployment algorithm MFMTP by dividing the optimization problem into graph partitioning problem.MFMTP defines a new metric called Relevancy Degree(RD)to quantize traffic dependency across network function instances and iteratively optimizes the initial deployment solution based on RD to minimize the amount of traffic between the switch fabric and the network function pool.Simulation results show that MFMTP outperforms the exsiting greedy algorithm in reducing interactive traffic and improving overall system throughput.To find out an efficient packet scheduling in SFCSA,this thesis modifies the FIRM algorithm that is widely used at present,and proposes a new scheduling algorighm BSC-FIRM.BSC-FIRM defines a new metric called Service Capacity(SC)to quantize the status of each queue during scheduling and optimizes the iterative scheduling of FIRM algorithm based on SC to achieve high-performance packet scheduling.Simulation results show that BSC-FIRM algorithm outperforms the comparative algorithms in terms of the delay and packet loss rate for different traffic models in SFCSA.