engleski

Relevance and Implications of Long-Memory Background Traffic on ABR Explicit-Rate Allocation: A Single Bottleneck Case

In this paper, relevance and implications of background VBR traffic correlation structure, pertaining to guaranteed services, on ABR explicit-rate allocation is discussed. Both, background traffic exhibiting long-range dependency (LRD) and short-range dependency (SRD) are considered. Three explicit-rate switch algorithms are analyzed: Explicit Rate Indication with Congestion Avoidance (ERICA+), Enhanced Distributed Explicit-rate Allocation Algorithm (EDERA), and Enhanced Proportional Rate Allocation (EPRCA). Dependence of the performance metrics, including average queue length, link utilization, and allocation fairness, on intensity of both LRD and SRD is evaluated, for a single bottleneck case. First two moments of background traffic are held fixed whereas the trace autocorrelation structure is varied. At the frame level, the background traffic is modeled as fractional ARIMA(p, d, q) process, synthesized by the FFT procedure proposed by [1]. From the obtained results, it could be generally concluded that uncorrelated background traffic gives conservative performance metrics, i.e.. presents the worst-case. In addition, it is shown that ABR rate-allocation in presence of LRD VBR background traffic is resulting in spreading of guaranteed traffic LRD to ABR traffic, referred as contagious effect.

Available Bit Rate; Explicit Rate Allocation; Self-similarity; Long-range dependence; Discrete-Event Simulation

nije evidentirano