Algorithmic foundations of the internet: roundup
CAAN'04 Proceedings of the First international conference on Combinatorial and Algorithmic Aspects of Networking
| Hi-index | 0.00 |
The Internet is currently composed of thousands of networks (autonomous systems) containing several hundred thousand routers. It increases in size at an exponential pace and in a decentralized, adhoc manner. This combination of size, rapid growth and decentralization poses tremendous challenges to an outside observer attempting to understand the internal behavior of the Internet. As the observer will have access to only a very small fraction of components in the interior of the network, it is crucial to develop methodologies for network tomography. i.e., identifying internal network performance characteristics based on end-to-end measurements. In this talk, we overview the nascent field of network tomography. We illustrate problems and issues with examples taken from our MINC (multicast inference of network characteristics) project [1]. Briefly, the MINC project relies on the use of end-to-end multicast loss and delay measurements for a sequence of active probes to infer loss and delay characteristics on the distribution tree between a sender and a set of receivers. Thus, given a set of potential senders and receivers, two ingredients are required. The first, statistical in nature, is a methodology for inferring the internal characteristics of the distribution tree for a single sender and set of receivers. The second, algorithmic in nature, is a methodology for choosing an optimal set of trees that can be used to infer the internal characteristics of the network. Last, we will attempt to identify challenging algorithmic problems throughout the talk and the rich set of avenues available for investigate the implementation and performance of different network tomography algorithms through experiments over the Internet.