Resource sharing in advance reservation agents

Abstract

We present an architecture where clients can make end‐to‐end resource reservations through agents. For each domain in the network, there is an agent responsible for immediate and advance admission control. Reservations from different sources to the same destination domain are aggregated as their paths merge toward the destination.

We show that network resources can be shared between immediate and advance reservations without being pre‐partitioned. Admission control for immediate reservations use information about resources to be allocated for advance reservations in the near future. An important parameter in the admission control algorithm is the so called lookahead time, i.e., the point at which we actually start making resources available for approaching advance reservations by rejecting immediate requests. In our model, preemption of immediate reservations is made in cases where the admission control cannot make resources available through rejection of immediate requests. The risk of preemption can be varied by changing the lookahead time. We explore, with simulations, the effects of providing advance reservations with this model. The results show the cost in terms of resource utilization, rejection probability and preemption probability.