Networks with multiple demand matrices

The classical approach for network design assumes a given single demand matrix. A single matrix is either based on demand estimations, traffic forecasts, population statistics, or simply generated randomly or manually. However, demands can never be known with precision. Traffic in networks might increase exponentially over longer time horizons but it also fluctuates heavily over the day or week.

To study such phenomena and to come up with reasonable approaches for multi-period, robust, or stochastic network design problems we also provide multiple demand matrices for some of our networks. These matrices are based on detailed measurements of traffic in real IP networks with a certain level of anonymization. Please consult the corresponding readmes.

The matrices on this site have gone through several steps of conversion, and some data was typed in by hand or transformed by automatic scripts. While we made a lot of effort to make no mistakes and to double-check the results, we cannot guarantee that everything is correct. Moreover, the original data may contain inaccuracies already.

We would like to thank Steve Uhlig, Yin Zhang, and the DFN Verein for providing us with real data. The data has been processed with help from Andreas Bley, Kai Hennig, Filip Idzikowski, and Christian Raack.

Please cite [3] and [4], when using the dynamic traffic data in your publications.

Network Granularity Horizon Number of matrices native ASCII TGZ XML TGZ
abilene 5 min 6 month 48096 tgz (46.9 Mb) tgz (51.3 Mb)
geant 15 min 4 month 11460 tgz (37.7 Mb) tgz (69.7 Mb)
nobel-germany 5 min 1 day 288 tgz (585 Kb) tgz (1.3 Mb)
1 day 1 month 28 tgz (66 Kb) tgz (133 Kb)
1 month 1 year 14 tgz (35 Kb) tgz (67 Kb)
germany50 5 min 1 day 288 tgz (5.5 Mb) tgz (6.2 Mb)
1 day 1 month 28 tgz (731 Kb) tgz (828 Kb)
1 month 1 year 14 tgz (387 Kb) tgz (439 Kb)
brain 1 min 7 days 9723 tgz (604.2 Mb) tgz (797.5 Mb)
1 h 375 days 8993 tgz (577.4 Mb) tgz (759.7 Mb)

© 2023 Poznan University of Technology (PUT)