Generative Matrix Completion for Real-Time Network Latency Estimation

Network latency estimation plays a critical role in network performance monitoring and management. However, with the escalating demand for real-time performance monitoring and rapid network adjustments in contemporary networks, existing latency estimation methodologies fall short in meeting the needs for instantaneous estimation. In this paper, we propose a generative matrix completion (GMC) scheme for real-time network latency estimation. First, we employ a novel matrix completion framework that leverages a pre-trained generative model as a structural proxy to capture the latency matrix lowrankness. The pre-trained generative model well learns the lowrank characteristics of latency matrices in the pre-training stage, and can map a condensed latent representation to the matrix space for instantaneous matrix completion. Secondly, we propose two tailored training strategies for the generative models to adapt to the diverse low-rank patterns observed in network latency data. Thirdly, to further expedite the estimation, a learned update rule is implemented to hasten the discovery of a suitable latent representation, culminating in the development of the GMC scheme. We also provide a theoretical recovery guarantee to reveal the error bound of GMC. Experimental results on real-world datasets show that the proposed scheme can achieve accurate latency estimation within 50 ms, while maintaining the relative square error of estimation at no more than 0.11 (as evidenced using the PlanetLab dataset).