Hierarchical virtual paths allocation in large-scale ATM networks using noncooperative game models

Abstract

Broadband Integrated Services Networks (B-ISDN) form the foundation for the modern communication networks due to their ability to transport different types of traffic classes such as voice, video and data. The Asynchronous Transfer Mode (ATM) networks were chosen to be the transmission mechanism of B-ISDN.

The configuration of Virtual Path (VP) connection services plays an important role in the design and operation of large-scale ATM networks. VPs can be regarded as "elastic bands" of reserved bandwidth on the network links. The major research challenge is to account for the fundamental trade-off between the overall network throughput and the processing load on the signaling system. The first being affected by the amount of reserved and hence not fully used capacity: while the increased signaling load is affected by the amount of connections to be processed. Therefore, it is essential to provide an algorithm for VP capacity allocation that achieves an optimal network operating point while guaranteeing the QoS at the call level and satisfies a priori bounds on the processing load of the call processors.

Recently, system-wide, centered control of large-scale networks was recognized to be prohibitive and impractical. A hierarchical structure segments a large network into smaller, manageable entities and provides more flexibility to the changes in the network topology. Techniques for representing and passing information between hierarchical levels are explored. An issue of accuracy of representation arises as the level of information abstraction increases due to the diminished quantity of details broadcast about the status of the network resources between the nodes.

Furthermore, users competing for the network resources can be regarded as a collection of noncooperative players acting in a greedy manner. For this purpose, game theory provides the analytical tools to model their behaviour. In general, a game with .V players, simultaneously deciding on their game strategy, can possess a steady state strategy profile, or equilibrium point known as the "Nash-equilibrium". A Nash-equilibrium is the operating point of the network from which a unilateral deviation does not help any player to improve his performance. Moreover, the Stackelberg game model is used to provide more control to network managers by allowing them to arbitrate the game in a prioritized manner. This is achieved by the process of associating higher priorities to connections generating higher revenue. The game proceeds then between connections of the same priority level, beginning with the highest priority connections deciding on their strategy profile and ending with the lowest priority ones responding by competing on the remaining network resources.

In summary, the objective of this work is to search for the best possible allocation of network resources among the competing users in such a way as to increase the network's revenue while maintaining the Quality of Service requirements of the users. Implementation wise, a coordination between local network managers is the framework within the network layers to achieve this objective. The subdivision of the optimization problem among smaller local managers becomes therefore computationally feasible.