The Empirical Core of the Multicommodity Flow Game Without Side Payments

Policy makers focus on stable strategies as the ones adopted by rational players. If there are many such solutions an important question is how to select amongst them. We study this question for the Multicommodity Flow Coalition Game, used to model cooperation between autonomous systems (AS) in the Internet. In short, strategies are flows in a capacitated network. The payoff to a node is the total flow which it terminates. Markakis-Saberi show this game is balanced and hence has a non-empty core by Scarf's Theorem. In the transferable utility (TU) version this gives a polynomial-time algorithm to find core elements, but for ASs side payments are not natural. Finding core elements in NTU games tends to be computationally more difficult. For this game, the only previous result gives a procedure to find a core element when the supply graph is a path. We extend their work with an algorithm, Incorporate, to which produces many different core elements. We use our algorithm to evaluate specific instances by generating many core vectors. We call these the Empirical Core for a game. We find that sampled core vectors are more consistent with respect to social welfare (SW) than for fairness (minimum payout). For SW they tend to do as well as the optimal linear program value $LP_{sw}$. In contrast, there is a larger range in fairness for the empirical core; the fairness values tend to be worse than the optimal fairness LP value $LP_{fair}$. We study this discrepancy in the setting of general graphs with single-sink demands. We give an algorithm which produces core vectors that simultaneously maximize SW and fairness. This leads to the following bicriteria result for general games. Given any core-producing algorithm and any $\lambda \in (0,1)$, one can produce an approximate core vector with fairness (resp. social welfare) at least $\lambda LP_{fair}$ (resp $(1-\lambda) LP_{sw}$).