The multi-budgeted and weighted bounded degree metric Steiner network problem

Abstract

We study the multi-budgeted version of the metric Survivable Network Design Problem (SND) [8], also known as Steiner Network problem, where besides the usual connectivity requirements (i.e., lower bound on the number edge-disjoint paths) between pre-specified pairs of points, we are also asked to satisfy a set of extra linear constraints (the budgets). Based on combinatorial properties of extreme point solutions of the natural LP relaxation of the problem and using the iterative method, we design new approximation algorithms with the following performance guarantee: if any edge participates in at most f ∈ N budgets, then we show how we can device an (f + 2, f + 2) bi-criteria approximation algorithm in polynomial time i.e., return a solution with cost at most (2 + f) times the optimal one with a potential budget overflow bounded from above by the same factor. In other words, the approximation and budget violation guarantee we achieve is a linear function of the maximum frequency among all the edges in the budgets, as opposed to a function of the number of budgets themselves and this constitutes a strict improvement over the previous approaches. Our approach is flexible enough and we demonstrate how it can be used to provide a (4, 4) bi-criteria algorithm for the Minimum Weighted Bounded Degree version of the SND problem in which we have the standard SND connectivity constraints and, additionally, for every vertex v in the graph we have an upper bound on the sum of the weights of the edges incident to that edge in any feasible solution. These problems, besides their obvious theoretical interest, have numerous real life applications from Cloud Computing to Optical Networking systems.