COPE: traffic engineering in dynamic networks
Proceedings of the 2006 conference on Applications, technologies, architectures, and protocols for computer communications
Finding hierarchical heavy hitters in network measurement system
Proceedings of the 2007 ACM symposium on Applied computing
R3: resilient routing reconfiguration
Proceedings of the ACM SIGCOMM 2010 conference
Multi-hour network planning based on domination between sets of traffic matrices
Computer Networks: The International Journal of Computer and Telecommunications Networking
Hi-index | 0.00 |
A traffic matrix represents the amount of traffic between origin and destination in a network. It has tremendous potential utility for many IP network engineering applications, such as network survivability analysis, traffic engineering, and capacity planning. Recent advances in traffic matrix estimation have enabled ISPs to measure traffic matrices continuously. Yet a major challenge remains towards achieving the full potential of traffic matrices. In practical networking applications, it is often inconvenient (if not infeasible) to deal with hundreds or thousands of measured traffic matrices. So it is highly desirable to be able to extract a small number of "critical" traffic matrices. Unfortunately, we are not aware of any good existing solutions to this problem (other than a few ad hoc heuristics). This seriously limits the applicability of traffic matrices. To bridge the gap between the measurement and the actual application of traffic matrices, we study the criticaltraffic matrices selection (CritMat) problem in this paper. We developed a mathematical problem formalization after identifying the key requirements and properties of CritMat in the context of network design and analysis. Our complexity analysis showed that CritMat is NP-hard. We then developed several clustering-based approximation algorithms to CritMat. We evaluated these algorithms using a large collection of real traffic matrices collected in AT&Týs North American backbone network. Our results demonstrated that these algorithms are very effective and that a small number (e.g., 12) of critical traffic matrices suffice to yield satisfactory performance.