Loss Tomography in General Topology

TL;DR

This paper develops and analyzes maximum likelihood and other estimators for loss tomography in general network topologies, providing theoretical properties and validation through simulations.

Contribution

It introduces a maximum likelihood estimator for general topologies and analyzes its statistical properties, advancing beyond prior simulation-based studies.

Findings

MLE is unbiased and efficient under certain conditions

The renewed MVWA removes the need to know variance beforehand

Simulation confirms the theoretical properties of the estimators

Abstract

Although there are a few works reported in the literature considering loss tomography in the general topology, there is few well established result since all of them rely either on simulations or on experiments that have many random factors affecting the outcome. To improve the situation, we address a number of issues in this paper that include a maximum likelihood estimator (MLE) for the general topology, the statistical properties of the MLE, the statistical properties of a frequently referred estimator called the moving variance and weighted average (MVWA), and a renewed MVWA that removes the restriction of knowing variance in advance from the MVWA. The statistical properties covers minimum-variance unbiasedness, efficiency, and variances of the estimates obtained by the estimators. Given the properties, we can evaluate the estimators without the need of simulations. To verify the properties, a simulation study is conducted that confirms the accuracy of the findings.