Estimating traffic parameters in queueing systems with local information

Abstract

In ATM networks a new connection is accepted if the CAC (call acceptance controller) at the entering node decides that the network has the requisite resources for the incoming traffic. For this, the CAC will need to monitor the network congestion based on the local information available to it. In this paper we propose several schemes that the CAC can use to estimate the traffic of the users on the various routes in the network by sending a ‘probing’ stream. First we consider a single node where the probing stream could be arbitrary while the total interfering traffic is a Poisson process. Knowing the mean service time of the interfering traffic and the delays experienced by the probing stream we provide algorithms to estimate the arrival rate of the interfering traffic. We model the node as a continuous time as well as a discrete time queue. Then we generalize the assumptions on the interfering traffic to make the overall traffic at the node a general stationary ergodic stream. For the Poisson interfering traffic we also consider a tandem of queues. Finally, for a product form Kelly network we provide estimates of the arrival rates of interfering traffic and the service rates of customers on the route of the probing stream of the CAC. These estimates are also used to provide the optimal ‘reshaping’ of the traffic stream of a user based on the delays experienced by the user in the network.