Performance analysis of a parallel link network with preemption

Preemption can be used to provide improved availability and reliability to high priority traffic on a congested network, or when the network experiences link or node failures and traffic needs to be rerouted. The use of preemption permits improved blocking probabilities and traffic alignment on shortest paths for high priority traffic at the expense of performance degradation for low priority traffic. Such policies have been deployed in a variety of scenarios and are very attractive for a service provider. For instance, the Internet Engineering Task Force (IETF) traffic engineering working group (tewg) pointed out the need for priority and preemption parameters as traffic engineering attributes in a multiprotocol label switching (MPLS) network. In this paper we analyze a simple two parallel link network supporting two service classes, where the high priority (HP) class has hard preemptive priority over the low priority (LP) class. This simple two link topology can be thought of as an abstraction of a (more desirable) primary route and a (less desirable) secondary route connecting a given source destination pair in a large network. The preemption policy permits both preemption from the primary link to the secondary link (a transfer) if possible, and eviction from either link if necessary. Analysis of the system is trivial for HP streams, and quality of service (QoS) is captured by the admission probabilities on the primary and secondary links (which are independent of the LP traffic). For the LP streams, however, the QoS is specified both by admission rates on both links as well as preemption rates from both links. We are able to analyze the preemption and departure rates of LP streams in this system in terms of the Erlang-B blocking probability equation for an M/M/c/c queue. Numerical and simulation results show very good agreement and demonstrate some interesting behavior.