Node Failure Localisation Problem for Load Balancing Dynamic Networks

TL;DR

This paper addresses the challenge of localizing node failures in load balancing, dynamically routed networks using a novel Markov Chain Monte Carlo approach, improving monitor placement strategies for better fault detection.

Contribution

It introduces a stochastic network model and applies MCMC inference to the NFL problem, extending network tomography to dynamic routing scenarios.

Findings

MCMC inference effectively localizes node failures in dynamic networks.

The proposed monitor placement algorithms outperform existing methods.

The study demonstrates improved fault detection accuracy in load balancing networks.

Abstract

Network tomography has been used as an approach to the Node Failure Localisation problem, whereby misbehaving subsets of nodes in a network are to be determined. Typically approaches in the literature assume a statically routed network, permitting linear algebraic arguments. In this work, a load balancing, dynamically routed network is studied, necessitating a stochastic representation of network dynamics. A network model was developed, permitting a novel application of Markov Chain Monte Carlo (MCMC) inference to the Node Failure Localisation (NFL) problem, and the assessment of monitor placement choices. Two nuanced monitor placement algorithms, including one designed for the NFL problem by Ma et al. 2014 were tested, with the published algorithm performing significantly better.