Delay-bounded rate adaptive shaper for TCP traffic in DiffServ Internet

Applications with different quality requirements set out the need for different Qualities of Service (QoS) to be provided in Internet. Differentiated Services (DiffServ) model is an architecture proposed to provide QoS in the Internet in a scalable way. Assured Forwarding Per Hop Behavior (AF PHB) is a QoS service class which provides a loss sensitive service. The DiffServ Service Provider (SP) delivers services to customers where traffic parameters are quantified in a Service Level Agreement (SLA). The incoming traffic from customers are policed in order to make sure that they meet the specifications in the SLA. The portion of traffic that is nonconformant with the SLA is not guaranteed to receive the service quality specified in the SLA. Shapers delay nonconformant packets in order to increase the ratio of traffic that is within the bounds specified in the SLA. If nonconformant traffic is tolerated in the SP network up to some extent, increasing the ratio of traffic that is complying with specifications in the SLA may lead to unnecessary delaying of packets and may decrease throughput. In this thesis, a shaper, called Delay-Bounded Rate-Adaptive Shaper (DBRAS), is introduced which tries to increase the ratio of traffic that conforms to the SLA while satisfying an upper-bound (Dmax) in the amount of delay it can apply to incoming packets (shaping delay). By avoiding unnecessarily large shaping delays, it is shown that throughput is increased. In order to have the shaper to adapt to changes in network topology, traffic, and different propagation delays, an adjustment algorithm is proposed where the shaper dynamically adjusts its Dmax value in order to increase throughput. The resulting shaper is called Dynamic DBRAS (D-DBRAS). The heuristic adjustment algorithm is greedy in that it adapts the maximum shaping delay in the direction where throughput increases. Results obtained from simulations show that throughput of TCP in AF PHB shaped by D-DBRAS can be increased by up to 65% compared with unshaped traffic. Simulations are performed in order to analyze effects of parameters such as propagation delay, buffer threshold levels, and offered traffic on the performance of D-DBRAS. It is also shown through simulations that by using the adjustment algorithm, the maximum shaping delay, Dmax, converges to regimes where throughput increases in response to changes in offered traffic.