On the impact of QoS-sensitivity on correlation structures

This paper assesses the feasibility of detecting the existence of network users who are dissatisfied with degradations in quality-of-service (QoS) through traffic data analysis. The dynamics of the number of streams on a link are described using a Markov chain, including human user actions of canceling data transfers in congestion. The correlation length, the time until the autocorrelation function drops to a specified value, is mathematically expressed. Numerical results show that the cancellations considerably decrease the ratio of two correlation lengths measured at different traffic loads under three conditions. First, the link capacity is small and/or every stream consumes a large amount of bandwidth. Second, the maximum number of data streams is large. Third, the stream lifetime is at least 10 times longer than the human response time.