Research On Software-Defined Traffic Statistics Model And Method

In a multi-tenant-based cloud computing network,with the rapid increase of network bandwidth and network traffic,the network becomes more and more complex.In order to be able to grasp the network status clearly and in real time,network measurement plays an important role.Network measurement can be used to better network management,such as virtual switch traffic statistics,congestion control,anomaly detection,etc.Network measurement also plays an important role in traffic engineering.Through network measurement,an end-to-end traffic matrix can be constructed.However,the existing multi-tenant-based cloud computing network has two challenges in network measurement:One is to realize flexible network measurement and low-overhead dynamic deployment.In a multi-tenant cloud computing network,tenants share the resources of measurement nodes,and multiple measurement nodes and multiple measurement tasks need to be dynamically deployed.Traditional networks use SNMP to manage the network.As the complexity of the network increases,the management overhead increases.And the dynamic deployment of measurement will take up a lot of bandwidth.Every time the network changes,the controller needs to recalculate the forwarding path and update the flow table.In the SDN network architecture,to realize the management configuration of multi-tenant,multiple messages need to be sent,and the management configuration is not flexible enough.The second is to achieve high precision of visualization of multi-tenant virtual network traffic.Multi-tenant network visualization puts forward higher requirements for network measurement algorithms.The measurement algorithm should further improve the processing speed of the algorithm and reduce the use of storage space while ensuring high accuracy.In addition,full consideration should be given to the resource allocation of the multi-tenant at the measurement node to meet the measurement needs of each tenant.The main work and innovations of this paper include:(1)Aiming at the problem of inflexible deployment and high cost of multi-tenant measurement tasks in the network,this paper proposes a software-defined traffic statistics framework based on SRoU.This framework supports dynamic enablement of the associated path,realizes the management of storage resources and computing resources of computing nodes,and reduces measurement deployment and processing overhead.The controller delivers the statistical results based on the SRoU collected packets to support multi-tenant traffic measurement and statistics tasks.(2)In order to realize the high-efficiency and high-precision statistics of multi-tenant virtual network traffic and realize network visualization,this paper proposes the DoubleHash traffic statistics algorithm with separation of cold flows and hot flows.Through the conservative update strategy oriented to the separation of cold flows and hot flows,the accuracy of cold flow statistics is improved;through the method of secondary hashing,the probability of hash collisions is further reduced.Compared with other Sketch-based traffic statistics algorithms,Double Hash can achieve the same accuracy as other algorithms in a smaller storage space.At the same time,a variety of algorithm optimization schemes are proposed to further improve the accuracy and performance of the algorithm.(3)This paper implements a prototype system based on the Ariane architecture of the RISCV instruction set,which verifies the software-defined traffic statistics framework based on SRoU and the function of the Double Hash algorithm.A measurement performance optimization scheme based on pre-hashing is proposed,which further improves the throughput of the Double Hash algorithm update data packet.In summary,this article has conducted an in-depth study on the traffic statistics in the network,and proposed a software-defined traffic statistics framework based on the SRoU,which reduces the cost of measurement deployment.The Double Hash algorithm designed with cold flows and hot flows separation can still maintain high measurement accuracy in a small storage space,and achieve the goal of high-precision network visualization.Finally,the software-defined framework and the Double Hash algorithm with separation of cold flows and hot flows are functionally verified on the prototype system.