Optimization And Application Of Flow Table Resource Management In Programmable Data Plane

The new network architecture software defined network(SDN),with the idea of separating the control plane from data plane,makes the network gradually programmable.The key to network programmability lies in the programmability of data plane.With the development of chip technology,the idea of programmable data plane(PDP)is becoming more and more realistic,but the high flexibility also brings some problems in resource management.In particular,as an important resource in the programmable data plane,the flow table resource management has ushered in new challenges with the improvement of the programmable ability of the data plane.Specifically,the PDP switch can support user-defined the width of flow table.If the resource of flow table in PDP switch is not managed properly,it will lead to serious fragmentation of flow table resources.Also,the problem of insufficient flow table space is not unique to PDP switches.Fast hardware storage resources have always been limited in switches.The high performance of PDP equipment can also greatly improve the processing performance of the virtual network function service chain(vNF-SC).Migrating the vNF on the service chain from the virtual machine(VM)to the PDP switch or SmartNIC,can not only reduce the processing latency of the vNF but also reduce the hops of the link.However,considering the actual network deployment,the high deployment cost of PDP equipment has to be included in the problem constraints.Under the cost constraint,the optimal management and utilization of programmable data plane resources can be realized by considering the deployment scheme of PDP equipment and the optimal deployment of vNF-SC.Generally,compared with the traditional SDN data plane,the programmable data plane has more efficient and flexible processing performance,but it suffers problems of flexible flow table width and shortage of flow table resources.Under the constraints of defects and shortboards,the high-performance advantages of PDP equipment are also difficult to benefit the overall network.Therefore,this dissertation is committed to the research on the flow table resource management and optimization of programmable data plane from three perspectives:repairing defects,making up shortcomings and exploiting advantages.The main innovations and contributions of this dissertation are summarized as follows:1.To solve the complex management problem of flow table width in programmable data plane with the network virtualization scene,this dissertation studies the flow table resource virtualization and network slicing technology in the physical network based on programmable data plane.Firstly,this work proposes an effective virtualization scheme of flow table resources to reduce the fragmentation of flow table resources.Then,this work discusses the virtual network embedding(VNE)problem based on the flow table virtualization scheme,and proposes a three-layer VNE problem to maximize the utilization of physical network resources in the programmable data plane.At the same time,this work establishes an integer linear programming model to accurately solve the three-level VNE problem,and designs an efficient heuristic algorithm to obtain the solution quickly.Finally,this work designs and implements a virtual network management system to verify the feasibility of the proposed solution.2.Aiming at the shortboard problem of insufficient flow table resources on PDP devices,this dissertation considers expanding the limited flow table resources on PDP switch.This work mainly solves two problems:one is how to realize the expansion and placement of flow table resources with low latency,so as to ensure that the remote flow table expansion scheme will not become a new performance bottleneck of programmable data plane.The second is how to minimize the cost of remote flow table resource expansion scheme.Specifically,this work first considers using PDP switch as the Top-of-Rock(ToR)switch in the data center network.Each ToR switch can use remote direct memory access(RDMA)to remotely store flow tables in multiple rack servers connected to it.Then,based on the design of flow table expansion technology,this work explores the remote flow table resource placement model on PDP switch to minimize the bandwidth and latency overhead introduced by RDMA operation.Then this work proposes a mixed integer linear programming model to solve the problem,and designs an approximate algorithm to solve it in polynomial time.Finally,this work also designs a prototype system and test the application in the real environment,so as to verify that the remote expansion of flow table resource is feasible.3.To exploit the high-performance advantages of PDP equipment,this dissertation considers balancing the high-performance and high deployment cost of PDP equipment in the real network scenario,and realizes the optimal deployment of vNF-SC based on programmable data plane through the optimization of programmable data plane resources.This work cooperatively considers the upgrading of PDP equipment and the optimal deployment of vNF-SC.Firstly,an integer linear programming model is established to represent the overall optimization problem.Then,according to the idea of aggregating service chain requests,a fast algorithm for data center network scenario is designed to speed up the problem-solving.After that,this work also considers a more universal fast algorithm,divides the overall problem into two steps,and designs the approximate algorithm respectively.A large number of simulation results show that algorithms proposed in this work can well organize the programmable data plane resources and exploit the high-performance characteristics of PDP equipment as much as possible.