Research On In-band Network Telemetry Technology Based On P4

With the vigorous development of the Internet,networks with higher network throughput,larger topology,and more fluctuating network traffic make operators more reliant on network monitoring.Traditional network measurement is not enough to adapt to such a huge network topology and complex network traffic,and the continuous improvement of software capabilities also brings more and more quantity and quality requirements to the supply of network measurement data,which makes it more and more urgent to realize a general-purpose,high-performance,and highly scalable in-band network telemetry system.However,the existing network telemetry solutions are still in the development stage,and some generalized data formats,detailed construction standards,and specific construction details have not been fully standardized.This paper studies the softwaredefined network architecture using P4 in the data plane to realize the in-band network telemetry system,and the main work is as follows:(1)The key components of the telemetry system,the main functions of different modules and the corresponding design schemes are analyzed,and finally a network telemetry system that supports dynamic expansion,high performance and general purpose is built.At the same time,it considers the importance of the data plane to realize event detection,and points out the dynamic expansion scheme of the telemetry collection system.After verification and construction of the design scheme,it is shown that the scheme has good data processing ability and efficient event detection and filtering mechanism under the premise of fulfilling the basic in-band telemetry requirements.(2)Aiming at the problems that the current link failure detection scheme occupies too much network bandwidth when high timeliness is required,is difficult to handle congestion and packet loss,and cannot directly locate link failures,the IPFD scheme for fast detection and positioning of link failures on the data plane is implemented based on P4.Under the premise of ensuring transparency,low intrusion,high efficiency,and generality,it can quickly detect and locate link failures at the granularity of tens of milliseconds.Simulations show that this scheme occupies less bandwidth and is insensitive to network congestion,and can directly locate link failures quickly without the participation of the control plane.(3)Aiming at the shortcoming that the fast recovery algorithm of the failed link may cause path congestion,and the link recovery that the control plane participates in takes too long,the fast route recovery scheme FR-RFMP is implemented based on P4 using multipath.Based on the multipath pre-delivered by the control plane,this scheme can quickly restore the link in the event of a link failure.It can evenly distribute traffic during operation,without the participation of the control plane,and implement data plane routing based on global congestion.Simulation shows that this scheme can effectively avoid network congestion,reduce the completion time of service flow,and has good performance compared with other fast route recovery methods.