Adaptive Scheduling for Improved Quality Differentiation

In this paper we compare the performance of static and adaptive provisioning methods with ns2-simulations. In the simulations we use a topology that captures the most fundamental aspects of real networks: multiple bottleneck links, paths with low and high delay, cross traffic, bidirectional traffic etc. We also use a representative traffic mix consisting of five different traffic types: FTP, HTTP, Video, VoIP and control traffic. For the static provisioning case we use capacity as a provisioning parameter and for the adaptive case we use packet delay. The scheduling algorithms that we investigate are Deficit Round Robin (DRR) and delay-bounded HPD (Hybrid Proportional Delay), which is our own version of the HPD algorithm. We test both algorithms with different load levels, application mixes, initial provisioning parameters and queue management methods (DropTail and RED). From the simulations we measure the end-to-end aggregate throughput, packet loss and delay for each traffic type. In addition, for web and ftp we measure object throughput and its variability to better capture the performance from the user’s point of view. According to our results the delay-bounded HPD algorithm is better able to achieve the targeted provisioning goal than the static DRR algorithm regardless of the load level, application mix or queue management method used. In this paper we also discuss how the measurement procedures of the delay-bounded HPD algorithm should be implemented in practice.