Research On Network Latency Prediction Algorithm Based On Network Coordinates

Network latency,as an important indicator of network performance,determines the scalability level of distributed network applications.However,the current complex and diverse network architectures have resulted in high cost associated with direct measurement of network latency.The use of network coordinate-based latency prediction algorithms can achieve efficient latency prediction with low overhead and easy scalability.This algorithm uses measured latency between some reference nodes to map all nodes into an Euclidean space,forming a Network Coordinate System(NCS).Through distance calculation based on coordinates,NCS can predict the overall network latency information.Currently,network latency prediction algorithms based on network coordinates sufer from problems such as measured latency pollution,suboptimal reference nodes,and rough coordinate error estimation,which result in poor performance in aspects such as accuracy of latency prediction and stability of node coordinates.Based on the above subject,the main research achievements and innovations are as follows:(1)Two algorithms for suppressing random network latency contamination are proposed:the AWB-Filter and WK-Filter,which preprocess measured contaminated latencies.The AWB-Filter uses time gradient and latency data gradient to jointly solve the weights of the latency within the window,and further incorporates a low-pass guide intermediate layer to promote the latency contamination suppression process.The WK-Filter introduces a gain function to solve the weight relationship between the measured latency and the output latency,transforming the latency contamination suppression process into a recursive solving process.Experiments show that both algorithms can suppress anomalous fluctuations with contaminated latencies and maintain the statistical characteristics of the original latencies.Additionally,the AWB-Filter improves NCS latency prediction accuracy by 10.8%,and the WK-Filter improves NCS node coordinate stability by 16.8%.(2)A reference node selection algorithm based on stage and cluster methods is proposed,which divides the reference nodes into nearby reference nodes and remote reference nodes.Through the node broadcast stage and node selection stage,nearby reference nodes are selected for the nodes,and through cluster division of nodes,a coordinate adjustment prediction strategy and cutoff mechanism are added to select remote reference nodes from different clusters.The reference node system composed of nearby and remote reference nodes serves as a reference for node coordinate adjustment to promote stable convergence of node coordinates effectively and improve NCS latency prediction performance.Experiments show that the reference node system proposed in this paper makes the convergence process smoother and can improve the accuracy of latency prediction by up to 44%and node coordinate stability by up to 26%.(3)A network latency prediction algorithm based on optimized coordinate error estimation called E-Vivaldi is proposed,and a node management scheme for NCS generated by E-Vivaldi is provided.E-Vivaldi introduces the RTT measurement error and average positioning error of reference nodes to optimize coordinate error estimation.It integrates random network latency pollution suppression algorithm and reference node system to make network latency prediction more efficient.The NCS node management scheme effectively addresses the problem of network node failure or disconnection by introducing member management protocols,heartbeat mechanisms,and update mechanisms.Experiments show that E-Vivaldi can improve the accuracy of NCS latency prediction by up to 57.5%and the stability of node coordinates by up to 31.9%.