End-to-End Queueing Delay in ATM Networks

Abstract

We focus on a single connection in an AIM network. To study delay for this connection, we adopt a discrete time model for several queues in tandem. A service availability process at each switch along the path of the connection indicates how many packets from the connection can be served in each slot at that switch. The service availability process at a switch is influenced by the presence of cross traffic as well as the underlying scheduling policy at the switch. The quality of service is succinctly described by the network service process. In particular, the end-to-end delay can be expressed as a simple non-recursive function of the network service process and the cell arrival process for the connection.

We consider the impact on end-to-end delay of constraining the scheduling policy at each switch along the connection according to a service curve. In particular, we develop the concept of a guaranteed network service curve, which may prove to be of some value in the service definition of AIM networks. Finally, utilizing a recently proposed stochastic model for the cell arrival process for the connection, as well as for the service availability processes, we present bounds on the probability distribution of end-to-end delay.