A Methodology for Virtual Network Partitioning: The Deterministic Approach

This paper introduces a methodology for virtual network partitioning in a single node with a generalized processor sharing scheduler. The approach is based on the worst-case service curve provisioning. The true service curve is a function of the number of users and their traffic fluctuations. In this paper, we define a deterministic universal service curve as a lower bound to the true service curve. The universal service curve is time-invariant and is independent of traffic characteristics. We design a novel call admission controller to guarantee that the true service curve is always lower bounded by the deterministic universal service curve. This is achieved by adapting a suitable regulator to the input traffic of each user and allocating appropriate scheduler parameters. The universal service curve is then used to quantify the performance of the network in terms of the maximum delay and the maximum backlog. We further use the universal service curve to propose a virtual subnetwork decomposition. The assignment of an input traffic to a certain subnetwork is based on the proximity of its upper envelope function to the universal service curve. We use min-plus algebra and network calculus to concoct our approach. By using min-plus algebra, we establish a duality between the maximum delay and the maximum backlog.