A study on delay performance of schedulers in aggregate traffic environments

Abstract

There are two major drivers for network services with QoS guarantees. One comes from applications with stringent QoS requirements. Two possible examples of such applications are IP telephony, video-on-demand, IP TV over the Internet. The other major driver for network QoS is the need for service differentiation due to competitive nature of the marketplace. In recent years the highlighted Differentiated Services (DiffServ), aggregate class-based traffic management and control have been used to mitigate scalability issues of Integrated Services (IntServ) in backbone networks. Fair Queueing schedulers can provide service isolation and guarantee QoS to all aggregate classes by reserving certain ratio of service-to-arrival rates. Although the GPS variants provide the network operator with a set of class differentiation parameters (i.e. the proportional bandwidth share of each class), the actual queueing delays at a bandwidth allocation server depend on the bandwidth share, and the input load and traffic burstiness of each class. Consequently, although the bandwidth differentiation is controllable, the short term delay guarantee is not. It is widely assumed that attempting to provide deterministic QoS guarantees for the aggregate traffic classes is difficult due to the unpredictable temporal and spatial variations in the aggregate traffic. First contribution of this dissertation is to guarantee absolute delay constraint in aggregate network. Although Fair Queueing can control the bandwidth differentiation, it can not guarantee absolute delay constraint because the forwarding resources allocated to each class do not follow the actual class load variations. We propose Absolute Delay Guaranteed Round Robin (ADRR) to support hard delay constraint by following the actual class load variations. Second contribution of this dissertation is to guarantee the short term delay while being able to attain long term fairness in aggregate QoS network. It is important to main...