On the Multi-Commodity Flow with convex objective function: Column-Generation approaches
The purpose of this work is to develop an algorithmic optimization approach for a capacitated Multi-Commodity flow problem, where the objective is to minimize the total link costs, where the cost of each arc increases convexly with its utilization. This objective is particularly relevant in telecommunication networks, where device performance can deteriorate significantly as the available bandwidth on a link becomes limited. By optimizing this convex function, traffic is efficiently distributed across the network, ensuring optimal use of available resources and preserving capacity for future demands. This paper describes the Convex Multi-Commodity Flow Problem and presents methodologies to solve both its Splittable and Unsplittable variants. In the Splittable version, flows can be fractionally distributed across multiple paths, while in the Unsplittable version, each commodity must be routed through a single path. Our approach employs Column-Generation techniques to address the convexly increasing cost functions associated with arc utilization, effectively accommodating various forms of convex increasing cost functions, including nondifferentiable or black-box convex increasing functions. The proposed methods demonstrate strong computational efficiency, offering a robust framework for managing network flows in complex telecommunication environments.