Resource Provisioning for a Multi-Layered Network

Given the growth, complexity, and size of the Internet, new methodologies are needed to support cost-effective resource provisioning. This paper provides a cost-based polynomial-time heuristic algorithm for resource provisioning optimization called multi-layered market algorithm (MMA). The MMA is solvable for multi-layered, multi-technology, and practical-sized networks, where the traffic is modeled as a combination of constant bit-rate and variable bit-rate (VBR) traffic streams. A VBR stream is modeled either by a Gaussian process or by a Poisson Pareto burst process (PPBP) which under certain parameter values is long-range dependent - a known characteristic of the Internet traffic streams. The consideration of VBR traffic models in a multi-layered network optimization is a key novel aspect of MMA. The MMA considers a range of transport technologies operating in layers and traffic sharing schemes. The MMA implements flow-size-based routing where flows according to their sizes are routed independently. As routing affects resource requirement, such considerations are important for resource provisioning by the given cost models. The complexity resulting from these considerations, including layering and PPBP traffic, requires a simplified design philosophy which in this paper, is based on adopting the shortest path routing in each layer. This is achieved by MMA which is based on an iterative algorithm, and resource provisioning that is performed link-by-link in all layers. As a benchmark for MMA, we provide an integer linear programming (ILP) formulation for a multi-layered network optimization problem with fixed end-to-end demands. The MMA is validated by comparing its solutions to those ILP results in different variants of a six-node network, and its software is verified using double-entry bookkeeping - a method commonly used in accounting systems. The MMA runs on a platform called network mark-up language, which enables visualization and further validation of the results.