DMCA
Optimization of Network Traffic Measurement: A Semidefinite Programming Approach
| Citations: | 6 - 4 self |
BibTeX
@MISC{Bouhtou_optimizationof,
author = {Mustapha Bouhtou and Stéphane Gaubert and Guillaume Sagnol},
title = {Optimization of Network Traffic Measurement: A Semidefinite Programming Approach},
year = {}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
We study the problem of traffic matrix estimation in large IP networks. The problem of inferring the traffic for each (origin- destination) pair in the network from the link measurements has been widely studied over the past 10 years. Recently, the possibility of using network-monitoring tools such as Netflow (cisco Systems) has renewed the interest for this problem, the aim being now to make the best possible use of such tools. Because of its expensive cost of use and deployment, optimizing the set of interfaces on which Netflow should be installed (and/or activated) is of first interest. We model the optimal Netflow deployment problem in terms of experimental design, which allows us to use convex programming and semidefinite programming techniques. The principle is to choose a set of localizations for Netflow that minimizes (in a certain sense) the variance of an estimator for the traffic matrix. We show that this problem is a generalization of the rank-maximization of the observation matrix, and that the greedy algorithm provides a (polynomial time) approximation of the solution by (1 − 1/e). We give experimental results on real networks and we show that an approach based on the rounding of a continuous relaxation improves on other approaches, including greedy algorithms.
Keyphrases
semidefinite programming approach network traffic measurement greedy algorithm traffic matrix estimation polynomial time continuous relaxation improves link measurement semidefinite programming technique certain sense origin destination possible use cisco system optimal netflow deployment problem large ip network expensive cost convex programming experimental design network-monitoring tool observation matrix first interest real network traffic matrix experimental result
