An online dynamic traffic matrix completion method in software defined networks

Abstract

Online temporal Traffic Matrix (TM) estimation is important for network management and traffic engineering. However, current estimation methods are insufficient in estimation accuracy and measurement cost. In this paper, by combining the low rank feature of the traffic matrix and the flow measurement capability in Software Defined Networks (SDN), we propose a novel online dynamic temporal traffic matrix completion mechanism DTMC. DTMC evaluates the impact of different Origin–Destination (OD) flows on improving the traffic matrix estimation accuracy from the perspective of uncertainty and temporal stability, selects an appropriate number of “most-informative” flows to construct the measurement set, and finally recovers the complete traffic matrix by using the partially measured OD flow information dynamically. The experiment results on two Internet measurement datasets show that DTMC can estimate the dynamic temporal traffic matrix accurately only by consuming a small amount of measurement resources.