Modeling Random Early Detection in a
Differentiated Services Network
Alhussein A. Abouzeid, Sumit Roy
An analytical framework for modeling a network of Random Early Detection (RED) queues with mixed traffic
types (e.g. TCP and UDP) is developed. Expressions for the steady state goodput for each flow and the average
queuing delay at each queue are derived. The framework is extended to include a class of RED queues that provides
differentiated services for flows with multiple classes. Finally, the analysis is validated against ns simulations for
a variety of RED network configurations where it is shown that the analytical results match with those of the
simulations within a mean error of 5%. Several new analytical results are obtained; TCP throughput formula for a
RED queue; TCP timeout formula for a RED queue and the fairness index for RED and Tail-Drop.
Random Early Detection, TCP, UDP, differentiated services, performance analysis, network modeling.
The diverse and changing nature of service requirements among Internet applications man-
dates a network architecture that is both flexible and capable of differentiating between the
needs of different applications. The traditional Internet architecture, however, offers best-effort
service to all traffic. In an attempt to enrich this service model, the Internet Engineering Task
Force (IETF) is considering a number of architectural extensions that permit service discrimi-