A scheduler for delay-based service differentiation among AF classes

In the Differentiated Services (DS) framework, service differentiation is performed among the aggregates (collection of one or more microflows) rather than among the microflows (data stream pertaining to a single connection). We analyse three quality metrics namely bandwidth, loss and delay (that might be used for defining a service differentiation at a DS node) on two criteria 1) service differentiation should be respected at all loads and 2) service provision at aggregate level should scale down to microflow level without being microflow aware. We find that bandwidth requires microflow aware management, loss lacks in simplicity (though it satisfies the criterion # 2), and delay is the right candidate. Ensuring better delays at an aggregate level also means ensuring better delays for all the included micro flows, and additionally it is easier to define a scheduler that can adapt itself to the relative loads of the aggregates so that relative delays between aggregate are preserved at all loads. Our objective is to provide relative quantification service in DiffServ by a delay-based scheduler while satisfying both criteria. Delay is also a meaningful QoS parameter for both interactive real-time applications and TCP applications, since the mean TCP throughput is roughly inversely proportional to the RTT. We, therefore, develop a scheduler for Assured Forwarding (AF) PHB where service differentiation among aggregates is based on delays. We provide simulation results that prove that relative delays among aggregates are perfectly respected at all loads.