Compressed sensing based network tomography using end-to-end path measurements

Abstract

Network tomography can help detect network link delay. End-to-end path measurement is a promising method in network tomography which could reduce measurement overhead. In end-to-end path measurement, however, deriving link states is computational complex due to the existence of massive links. Because only a few links are congested at the same time, we can locate sparsely congested links by using compressed sensing (CS). In this paper, based on expander graphs, we demonstrate that routing matrix constructed by full tree network can satisfy Restricted Isometry Property (RIP) and be used as measurement matrix. Finally, sparsely congested links are given with CS. Evaluation results show that routing matrix constructed in this paper can help achieve high link delay recovery accuracy, i.e., small recovery error by using 1i-minimization in CS. It also illustrate that the recovery error decreases with the number of congested links decrease.