Indirect Estimation of Link Delays by Directly Observing a Triplet of Network Metrics

Abstract

This paper presents an improved indirect estimation link delays from a triplet of network metrics; path delays, packet loss rate (PLR), and jitter by using indirect inverse modeling techniques. conventionally a network metric is estimated by directly observing another network parameter. Based on the evidence in the literature that path delays, PLR, and jitter are interdependent, this work exploits this mutual interdependent of this triplet of metrics based on the notion that a better observation leads to better estimation. We applied NTF1 model, a variation of non negative tensor factorization (NTF) for this purpose and estimated link delay from a triplet of metrics. Evaluation process used data from an experimental test bed that consists of standard networking devices. The estimated link delays were correlated to actual link delays to benchmark the accuracy of estimation. Results showed a better correlation between the estimated and measured link delays when a triplet of metrics is used.