IP Network Performance Anomaly Multipath Parallel Detection And Localization Algorithm

IP network anomaly detection and localization is critical in ensuring a high network quality of service. In this paper, we introduce a large scale network performance anomaly detection and localization algorithm (LSNA) based on active probing, aiming at avoiding waste of unnecessary probes and reducing detecting time by decreasing detection cost and probe selecting rounds, while lessen link load within the network. We propose a strategy that sends a number of probes (k) within each time interval in a multi anomaly link network during its detection phase. Also we experimented over12different network topologies and research on the relationship between the number of probes send within each time interval (k) and the number of probes in the network (Np) in order to find the optimum k in the algorithm implementation. In localization phase, we avoid several types of useless probes to enhance the localization efficiency and save probing messages. Also we develop a distance based assessment method to evaluate the anomaly probabilities of the unidentified anomaly state links. Finally we simulate two former representative algorithms and compare with our improved method LSNA on these different network topologies, the results show that in detection phase our improved algorithm outperforms the former algorithms in saving detection phase probing cost by18%and minimize link load by33%with Overlap Allowed strategy. In localization phase, our method has saved5%probing cost with localization accuracy exceeding over92%.