Long Tail Behavior of Queue Lengths in Broadband Networks: Tsallis Entropy Framework

A maximum entropy framework based on Tsallis entropy is proposed to depict long tail behavior of queue lengths in broadband networks. Queue length expression as measured in terms of number of packets involves Hurwitz-zeta function. When the entropy parameter q in Tsallis entropy is less than unity, the distribution of packets yields power law behavior. In the limit q tending to unity, Tsallis entropy expression reduces to one due to Shannon and well-known results of M/M/1 queuing system are recovered. Relationship between Tsallis entropy parameter q and Hurst parameter H (measure of self-similarity) is postulated. A numerical procedure based on Newton-Raphson method is outlined to compute Lagrange’s parameter β. Various relationships between traffic intensity ρ and Lagrange’s parameter β are examined using data generated from mean number of packets from storage model due to Norros. It is found that best fit corresponds to ρ being a linear combination of decaying exponential and power exponent in β for different values of entropy parameter q. Explicit expression for the probability that queue size exceeds a certain value is derived and it is established that it asymptotically follows power law for q less than one. The system utilization shows an interesting behavior when the parameter ρ is varied. It attains lower values than that of M/M/1 system for smaller values of ρ whereas situation reverses for higher values of ρ.