Programmable Telemetry Based Network Tomography Methods And Simulations

With the rapid increase in the variety and quantity of Internet services,users are demanding higher and higher quality of network services.How to optimize network performance is a challenge for network operators and network service providers.Only on the basis of a full understanding of network status and performance can network management and operation and maintenance be performed more effectively and rationally.Network Pivot can effectively use a small amount of network measurement data and apply network pivot algorithms to reason about the network performance that cannot be measured,and indirectly obtain the performance parameters inside the network.The emergence of data plane programmable technology brings great convenience to network measurement and management,and the use of in-band network telemetry on the programmable data plane enables more data needed for network tomography.Therefore,the study of network tomography methods and simulations based on programmable telemetry has important theoretical significance and application value.Network tomography reduces the measurement overhead and complexity by using easier to measure network data to infer the intrinsic performance parameters of the network.The main research contents and innovative work of this thesis are as follows:(1)The packet loss telemetry performance of two different alternatively labeled packet loss telemetry methods in network telemetry is studied,and the PMLE(PaketLoss Maximum Likelihood Estimation)packet loss inference algorithm is used to infer the internal performance of multicast and unicast networks,respectively,and the packet loss telemetry technique is used to improve the network packet loss tomography The method of improving the accuracy of network packet loss rate tomography using packet loss telemetry is proposed,and the telemetry nodes are deployed to the dominant nodes of the measurement network to reason about the network internal packet loss rate by combining telemetry data,which effectively solves the problem of error accumulation in packet loss rate tomography.(2)In the study of network delay tomography,the time delay between network nodes is measured by using probe packets carrying time stamps,and the internal link delay is calculated by using the network delay distribution inference algorithm based on moment estimation,and a method is proposed to improve the accuracy of network delay tomography by using the delay telemetry technology,deploying telemetry nodes to the dominant nodes of the measurement network,and using the telemetry delay data to infer the internal delay data for correction,effectively solving the inference error accumulation problem based on moment estimation.(3)Based on the theoretical and algorithmic studies,this thesis also conducts simulation validation of network packet loss rate tomography and network delay tomography based on programmable switches.The switch is programmed according to the different requirements of the experiments,and the corresponding network environment is built to verify the theoretical methods.The performance characteristics of the above measurement algorithms and measurement methods are summarized,as well as the final results of the network internal performance tomography.