Research On Key Technology For Packet Loss Estimation Based On Passive Measurement

Network performance measurement is an important content of network performance management.With the continuous increase of network services and related businesses,to be “informed control” about the operating status of the network,network performance management needs to shift from traditional port-based extensive management to refined management based on network paths and oriented to network flows.However,in the face of massive data and complex access network environment,how to effectively construct benchmark metrics in combination with measurement target,and under the premise of considering cost,accuracy,real-time,and realistic-usability,focus on investigating and distinguishing user behaviors in the management domain and external domain,so as to provide network performance measurements for different granularities is still fine unresolved issues in the research of current refined network performance management.In addition,in the face of information asymmetry in the current non neutral network environment,the effective evaluation and measurement of network performance is not only a concern of network managers,but also a part of the network user’s goals as an effective means to balance information asymmetry and arm end users in the current non-neutral network environment.Correspondingly,how to construct effective metrics to characterize the impact of path quality changes on end-user QoE,and use terminal data to accurately estimate related performance metrics is also a burning issue that needs to be solved in this field.As a connection-oriented and reliable transmission protocol,TCP’s congestion control mechanism has a natural response to changes in network performance,so it is a research hotspot in current academic circles to evaluate network performance with TCP streams.To this end,around the above issues and goals,this paper takes the important network performance measurement packet loss rate as the starting point,and mainly takes the passive measurement method to study the following contents for TCP flow:1.Establishment of TCP flow benchmark data: Currently,the benchmark data required for verifing the packet loss measurement at the border of the access network is obtained mainly through simulation.Even with measured data,the impact of capture packet losses and IP fragmentation was not considered in the process of constructing the benchmark packet loss rate.Therefore,relying on the IPTAS platform of CERNET Northeast China Network Center,this paper first designs and implements a considerable scale of controlled transmission to obtain the benchmark data needed for research.In the process of using the benchmark data to form the benchmark metric,the impact of capture packet losses and IP fragmentation is also considered to obtain a more accurate benchmark packet loss rate metric that can be used for network performance measurement research.2.Packet loss measurement at access network boundary based on Trace data: The advantage of packet loss measurement at access network boundary based on Trace data is the high accuracy,which depends on being able to effectively eliminate the dynamic influence of packet transmission along the network path.For this reason,on the basis of analyzing and demonstrating the feasibility of using ACK flow to measure packet losses at access network boundary,this paper constructs heuristic rules to utilize ACK flow to eliminate the impact of packet dynamics to improve measurement accuracy.On this basis,the realistic-usability of the proposed method in actual network applications is also considered to further strengthen its robustness.3.Packet loss measurement at access network boundary based on sampled flow records: Considering the burst and volatile nature of packet loss,to better meet the real-time requirements of current high-speed backbone network performance monitoring,on the premise of considering cost and real-time,this paper proposes an access network boundary packet loss measurement model LBTF based on sampled flow records.The model overcomes the problems that the previous estimation can’t distinguish the front and back packet loss of the monitoring point,the deployment difficulties and the consistency of cooperative operation.With the help of Mathis equation and regression analysis,the measurement of the front and back packet loss of the monitoring point can be realized when only the sampling stream records generated by the access network boundary equipment are used.4.Application-level packet loss estimation model based on TCP self-clocking mechanism: Although application-level packet loss estimation is simple and non-intrusive,the available traffic information related to packet loss estimation in application layer is coarse-grained.Therefore,there is still a gap between the accuracy of estimation and the actual application requirements.In order to solve this problem,this paper takes the L-Rex model as the carrier and uses the TCP self-clocking mechanism to mine the traffic information implied in the burst gap traffic pattern to improve the accuracy of application-level packet loss estimation.At the same time,for the sake of robustness,this paper further eliminates the adverse impact of packet reordering on packet loss estimation by refining retransmited packets.5.Useful packet loss estimation algorithm based on network layer measurement: Facing the information asymmetry in the current non-neutral network environment,the premise of improving end-user’s knowledge level is to construct appropriate metrics to effectively characterize the impact of path quality changes on end-user QoE.In this regard,application level packet loss rate is considered to be an appropriate metric.However,the accuracy of application-level packet loss estimation built on limited available resources is a challenge.Although the relevant research at home and abroad is devoted to improving the accuracy of application level packet loss measurement,the magnitude of improvement is limited.To this end,this paper attempts to utilize the rich information available at the bottom layer(network layer)to construct a new metric UTLR to effectively characterize the impact of path quality changes on end-user QoE.At the same time,the abundant available information at the network layer can be used to achieve an accurate estimation for UTLR.