End-to-end delay analysis for real-time networks

This paper presents an application of real-time queueing network theory to a particular network which models a video-on-demand server. Packets from each stream have stochastic arrival patterns, computation times and end-to-end delay requirements. We derive closed form solutions for the deadline miss rate of the packets. This methodology can be used to design admission control policies that provide statistical quality of service guarantees. By considering the actual requirements rather than worst-case requirements, the realtime queueing network approach significantly increases the server, utilization. This paper, illustrates how realtime, queueing network theory can be used to accurately predict the behavior of real-time systems in heavy traffic conditions. It shows how one can calculate the fraction of tasks that will miss their end-to-end deadlines. It presents new results on product form equilibrium distributions for multidimensional reflected Brownian motion processes when nodes are scheduled using EDF. Lastly, it presents simulation results to illustrate the excellent accuracy of the real-time queueing network approach and how the methodology can be used to provide statistical QoS guarantees.