Network Latency Estimation: A Tensor-based Weighted Singular Value Thresholding Method

Weice Sun, Jun Tao, Yifan Xu, Yang Gao, Zuyan Wang

Abstract: Network latency estimation has drawn much attention in recent years due to the growing number of personal devices and latency-sensitive applications with increasing user experience requirements. Nevertheless, it is difficult or even impractical to obtain the latencies among all nodes in a large-scale network because of the complicated network environment and unstable communication conditions. Therefore, it is increasingly critical to infer the latency of the whole network from partial measurements. However, existing works on network latency estimation are difficult to attain satisfactory accuracy due to the severe and time-varying network conditions and the unknown ranks of latency matrices. In this paper, we propose a novel weighted singular value thresholding approach based on the tensor model, which treats the singular values differently to achieve high estimation accuracy. Furthermore, we conduct extensive simulations on both synthetic and real-world latency datasets to demonstrate the efficiency of the proposed algorithm. Simulation results show that our algorithm can achieve significantly better performance in terms of estimation accuracy and computation time compared with the existing approaches.