Traffic matrix estimation using matrix-CUR decomposition

Traffic Matrices (TMs) are the primary input for a multitude of network operations and management activities within a backbone network. TMs can be collected directly using monitoring tools but with an additional communication and processing overhead. Thus, TM estimation techniques gained prominence, where the objective is to obtain an estimate of the TM using easily available information without any additional overhead. Principal Component Analysis (PCA) and its variants have been extensively employed for TM estimation. This paper demonstrates the sensitivity of PCA towards the rank parameter for TM estimation. To overcome this limitation, this paper proposes a CUR decomposition -based technique for traffic matrix estimation. Experimental results on real world traffic matrices collected from Abilene backbone network show that the CUR -based estimation technique exhibits negligible sensitivity to the rank parameter for estimating traffic matrices, and also, attains low estimation error in comparison to the existing schemes.