Long-Range Dependence of Internet Traffic Aggregates

Abstract

Abstract. This paper studies and discusses the presence of LRD in network traffic after classifying flows into traffic aggregates. Following DiffServ architecture principles, generic QoS application requirements and the transport protocol in use, a classification criterion of Internet traffic is established. Using fractal theory, the resulting traffic classes are analysed. The Hurst parameter is estimated and used as a measure of traffic burstiness and LRD in each traffic class. The traffic volume per class and per interface is also measured. The study uses real traffic traces collected at a University of Minho major backbone router in different periods of network activity.

Distributed Admission Control for QoS and SLS Management

“…Taking advantage of the consensual need for on-line QoS and SLS monitoring in CoS networks, the AC model makes admission decisions resorting to edge-to-edge measures of relevant QoS parameters for each service type and to SLS utilization. Thus, in this process, four main areas are interrelated (see AC decision criteria, which involves the establishment of service dependent AC equations; and (iv) CoS traffic characterization, which provides the knowledge of the statistical properties of the classes in the domain as a result of aggregation [16]. The use of policy-based network management and security considerations were left for further…”

“…The statistical properties of traffic aggregated into classes [16] need to be considered so that proper thresholds or safety margins to AC can be established. For instance, classes which exhibit long-range dependence may need large safety margins as this property has a significant impact on queuing behavior and on the nature of congestion, leading to unexpected QoS degradation.…”

Managing Services Quality through Admission Control and Active Monitoring

“…A differentiated services architecture needs to be supported by an adequate traffic classification strategy [10,11]. As initial policy, we treat TCP and UDP traffic separately, being UDP traffic further divided according to the applications QoS requirements stringiness.…”

“…Generically, three source models have been considered: Constant Bit Rate (CBR) sources, Exponential on-off (EXP) and Pareto on-off (PAR) sources. PAR sources with 1 < α < 2 under aggregation will allow to generate traffic exhibiting longrange dependence [10]. As this property has a significant impact on queuing behavior and on the nature of congestion leading to unexpected QoS degradation, larger safety margins may be needed.…”

A distributed admission control model for CoS networks using QoS and SLS monitoring

“…The statistical properties of traffic when aggregated into classes [20] need to be considered so that proper thresholds or safety margins to AC can be established. For instance, classes which exhibit long-range dependence may need large safety margins as this property has a significant impact on queuing behavior and on the nature of congestion, leading to unexpected QoS degradation.…”