Self-Similarity of AQM Filtered Traffic

Self-similarity of network traffic is a well-known paradigm for more than a decade, now. Following the first realization of this property, it has been a popular issue to question if the self-similarity persisted with the changing network architectures and conditions. However, the effects of Active Queue Management (AQM) mechanisms over the self-similarity of network traffic have almost been left uninvestigated. AQM mechanisms are developed to aid congestion control over the network. Recently, they are applied to commercial routers and they have become notable alternatives to droptail queues. It is easy to foretell that their priority aware variants, e.g. RIO, will find application in heterogeneous networks in a nearby future. The main principle of all AQM mechanisms relies on the adaptive congestion control mechanism of Transport Control Protocol (TCP). TCP sources adapt their sending rates to network conditions by decreasing packet transmission rates during congestion. AQM mechanisms notify congestion to TCP sources more effectively than droptail queues where they start dropping packets before the queue actually overflows, in a way taking a precaution before “the disaster happens”. Obviously their different loss pattern will change the statistical properties of the traffic and the degree of selfsimilarity. For this reason we find the joint behavior of TCP and AQM mechanisms worth investigating. In this work, we run simulations in NS-2 to analyze the impacts of queue management mechanisms; droptail, RED and BLUE on the self-similarity of network traffic. The degree of self-similarity is measured by the Hurst parameter (H). We estimate H with three well known methods: Aggregated Variances, Higuchi and R/S methods. We additionally analyze the throughput, goodput, queuing delay and loss rate measurements for these queue management techniques. It is observed that the H estimates are correlated with throughput, goodput, queuing delay and loss rate. We show that droptail RED and BLUE yield different degrees of self-similarity. Furthermore, we examine RED and BLUE under different parameter settings and show that the parameter selection of AQM mechanisms affects the H estimates.