Exploiting packet-sampling measurements for traffic characterization and classification

摘要: The use of packet sampling for traffic measurement has become mandatory network operators to cope with the huge amount data transmitted in today's networks, powered by increasingly faster transmission technologies. Therefore, many networking tasks must already deal such reduced data, more available but less rich information. In this work we assess impact on various monitoring-activities, a particular focus characterization and classification. We process an extremely heterogeneous dataset composed four packet-level traces (representative different access technologies operational environments) monitor able apply policies rates extract several features both aggregated per-flow fashion, providing empirical evidences First, analyze feature distortion, quantified means two statistical metrics: most appear deteriorated under low step, no matter policy, while only few remain consistent harsh conditions, which may even cause some artifacts, undermining correctness measurements. Second, evaluate performance classification sampling. information content features, though deteriorated, still allows good accuracy, provided that classifier is trained obtained at same rate target data. accuracy also due thoughtful choice smart policy biases towards packets carrying useful Copyright © 2012 John Wiley & Sons, Ltd.