Performance Impact of Wireless Mesh Networks with Mining Traffic Patterns

With mining traffic patterns, we evaluate the performance impact on wireless mesh networks. Genuine traffic traces are collected from the wireless mesh networks testbed, which tends to exhibit long-range dependent behavior under several Hurst index estimators. We analyze traffic traces and use clustering techniques to characterize patterns of individual users' behavior. After extracting traffic data from the raw data logs, we identify session clusters by employing the AutoClass tool and the K-means algorithm. Modeling and simulation were performed using the NS-2 tool. Based on the identified session clusters, we introduce source model based on wavelet. Simulation results indicate that traffic traces, compared to traditional traffic models, predict longer queues and, thus, require larger buffers in the network dimensioning.