A New Approach to the Control of Real-Time Traffic in Packet Switching Data Networks

by

Norival Figueira
Doctor of Philosophy in Computer Science
University of California, San Diego, 1995
Professor Joseph Pasquale, Chair

Abstract

This dissertation proposes a new approach to the control of real-time traffic in connection-oriented (i.e. fixed-routed) packet-switched data networks in the context of rate-based flow control. The general (i.e. old) approach is based on the characterization of sessions (i.e. sources of real-time traffic) according to some pre-defined traffic characterization model, which is generally expressed in terms of upper bounds (or lower bounds) on traffic measures such as interarrival times of packets, peak rate, etc. In the old approach, a network uses the traffic characterization of a session to allocate network resources to the session or to express the service guarantees of the session, i.e. service guarantees are expressed as functions of the parameters of the traffic characterization of the session.

My thesis is that it is possible for a service discipline to provide service guarantees that are completely determinable by a session's dynamic traffic behavior and to make efficient use of network resources. A new approach is proposed in which the network provides a session with service guarantees that are expressed as functions of the dynamic traffic behavior of the session, and no specific traffic characterization model is imposed or needed. This implies a major distinction between the new approach and the old one, since the former does not conceal the dynamic traffic behavior of sessions from the calculus of service guarantees (we will show that this allows a service discipline to provide upper bounds on the probability distribution of some service measures), while the latter can only provide service guarantees that are expressed as upper bounds on service measures.

To show that the new approach is feasible, a new service discipline called Leave-in-Time is developed and analyzed. The Leave-in-Time service discipline is able to provide a session with service guarantees that are completely determined by the dynamic traffic behavior of the session.

This dissertation also shows that leaky bucket constrained sessions are the only type of sessions for which an upper bound on delay can be provided by servers with an upper bound on link capacity. This result implies that the upper bound on delay provided by the Leave-in-Time service discipline is near-optimum.

Finally, implementation issues are discussed, and some simulations are performed to exemplify some of the service guarantees provided by the Leave-in-Time service discipline.


Last updated: Tue Oct 29 14:10:49 1996