Explicit Estimators for Loss Tomography

TL;DR

This paper introduces explicit estimators based on composite likelihood for loss tomography, offering comparable accuracy to full likelihood methods but with reduced computational complexity, applicable to both tree and general network topologies.

Contribution

The paper proposes a new composite likelihood approach that yields explicit estimators for loss tomography, simplifying computation while maintaining accuracy.

Findings

Explicit estimators perform nearly as well as full likelihood estimators in accuracy.

The explicit estimators are computationally more efficient than full likelihood estimators.

Methodology is applicable beyond tree topologies to general network structures.

Abstract

Full likelihood has been widely used in loss tomography because most believe it can produce accurate estimates although the full likelihood estimators proposed so far are complex in structure and expensive in execution. We in this paper advocate a different likelihood called composite likelihood to replace the full likelihood in loss tomography for simplicity and accuracy. Using the proposed likelihood, we propose a number of explicit estimators with statistical analysis. The analysis shows all of the explicit estimators perform almost as good as the full likelihood one in terms of accuracy and better than the full likelihood one in computational complexity. Although the discussion is restricted to the tree topology, the methodology proposed here is also applicable to a network of a general topology.