Leveraging router programmability for traffic matrix computation

Abstract

Traffic matrices are used by Internet Service Providers as an input for many business-critical activities like capacity planning and traffic engineering. Despite their importance, current approaches for computing traffic matrices are either unpractical or not arbitrarily accurate. In this paper, we propose a novel distributed architecture in which programmable routers autonomously compute parts of the whole traffic matrix. Our proposal conjugates router programmability with some ideas firstly proposed by Varghese and Estan [17] and avoids the need for packet sampling exploiting highly optimized counting mechanisms, already available on commercial routers. We describe our architecture and analyze main implementative issues. Also, we show the viability of our solution by describing a prototypical implementation and some experimental results.