Capacitated Network Design Games on a Generalized Fair Allocation Model

TL;DR

This paper studies a generalized model of cost-sharing network design games that accounts for overhead costs, analyzing the impact on efficiency measures like Price of Anarchy and Price of Stability.

Contribution

It introduces a broad class of cost-sharing functions with overhead and derives bounds on efficiency metrics, extending previous models without overhead.

Findings

PoA and PoS bounds remain tight under general cost-sharing functions.

PoS lower bound increases from log n to nearly n+1/n-1 with overhead.

Results apply to various realistic cost-sharing scenarios.

Abstract

The cost-sharing connection game is a variant of routing games on a network. In this model, given a directed graph with edge costs and edge capacities, each agent wants to construct a path from a source to a sink with low cost. The users share the cost of each edge based on a cost-sharing function. One of the simple cost-sharing functions is defined as the cost divided by the number of users. Most of the previous papers about cost-sharing connection games addressed this cost-sharing function. It models an ideal setting where no overhead arises when people share things, though it might be quite rare in real life; it is more realistic to consider the setting that the cost paid by an agent is the original cost per the number of agents using the edge plus the overhead. In this paper, we model the more realistic scenario of cost-sharing connection games by generalizing cost-sharing functions. The arguments on the model are based on not concrete cost-sharing functions but cost-sharing functions under a reasonable scheme; they are applicable for a broad class of cost-sharing functions satisfying the following natural properties: they are (1) non-increasing, (2) lower bounded by the original cost per the number of the agents, and (3) upper bounded by the original cost, which enables to represent various scenarios of cost-sharing. We investigate the Price of Anarchy (PoA) and the Price of Stability (PoS) under sum-cost and max-cost criteria with the generalized cost-sharing function. Despite the generalization, we obtain the same tight bounds of PoA and PoS as the cost-sharing with no overhead except PoS under sum-cost. Moreover, for the sum-cost case, the lower bound on PoS increases from $\log n$ to $n+1/n-1$ by the generalization, which is also almost tight because the upper bound is $n$.