Taking A Free Ride for Routing Topology Inference in Peer-to-Peer Networks

TL;DR

This paper introduces a passive, probe-free network tomography method for P2P networks that infers routing topology with high accuracy by analyzing end-to-end delay correlations, eliminating the need for cooperation from routers.

Contribution

It presents a novel passive approach to network topology inference in P2P networks using delay correlation estimation, avoiding active probing.

Findings

Achieved 92% accuracy in real-world Internet tests.

Simulations show about 95% accuracy in large-scale networks.

Method requires no synchronization or cooperation from routers.

Abstract

A Peer-to-Peer (P2P) network can boost its performance if peers are provided with underlying network-layer routing topology. The task of inferring the network-layer routing topology and link performance from an end host to a set of other hosts is termed as network tomography, and it normally requires host computers to send probing messages. We design a passive network tomography method that does not requires any probing messages and takes a free ride over data flows in P2P networks. It infers routing topology based on end-to-end delay correlation estimation (DCE) without requiring any synchronization or cooperation from the intermediate routers. We implement and test our method in the real world Internet environment and achieved the accuracy of 92% in topology recovery. We also perform extensive simulation in OMNet++ to evaluate its performance over large scale networks, showing that its topology recovery accuracy is about 95% for large networks.