Network Creation Game A. Fabrikant, A. Luthra, E. Maneva, C. H. Papadimitriou, and S. Shenker, PODC 2003 (Part of the Slides are taken from Alex Fabrikant’s presentation) 1 Context U C B E R K E L E Y C O M P U T E R S C I E N C E The internet has over 20,000 Autonomous Systems (AS) Every AS picks their own peers to speed-up routing or minimize cost 2 Question: What is the performance penalty in terms of the poor network structure resulting from selfish users creating the network, without centralized control? Goal of the paper U C B E R K E L E Y C O M P U T E R S C I E N C E Introduces a simple model of network creation by selfish agents Briefly reviews game-theoretic concepts Computes the price of anarchy for different cost functions 4 A Simple Model for constructing G U C B E R K E L E Y C O M P U T E R S C I E N C E N agents, each represented by a vertex and can buy (undirected) links to a set of others (si) One agent buys a link, but anyone can use it Cost to agent: Distance from i to j Pay $a for each link you buy Pay $1 for every hop to every node (a may depend on n) 5 Example U C B E R K E L E Y C O M P U T E R S C I E N C E 2 -1 3 -3 4 1 +a 2 1 a c(i)=a+13 c(i)=2a+9 (Convention: arrow from the node buying the link) 6 Definitions U C B E R K E L E Y C O M P U T E R S C I E N C E V={1..n} set of players A strategy for v is a set of vertices Sv V\{v}, such that v creates an edge to every w Sv. G(S)=(V,E) is the resulting graph given a combination of strategies S=(S1,..,Sn), V set of players / nodes and E the laid edges. Social optimum: A central administrator’s approach to combining strategies and minimizing the the total cost (social cost) It may not be liked by every node. Social cost: C(G) ci a E dG (i, j ) i i, j 7 Definitions: Nash Equilibria U C B E R K E E Y C O M P U T E R S C I E N C E Nash equilibrium: a situation such that no single player can unilaterally modify its strategy and lower its cost L Presumes complete rationality and knowledge on behalf of each agent Nash Equilibrium is not guaranteed to exist, but they do for our model The private cost of player i under s: C (i, S ) a Sv dG (i, j ) j 8 Definitions: Nash Equilibria U C B E R K E L E Y C O M P U T E R S C I E N C E A combination of strategies S forms Nash equilibrium, if for any player i and every other strategy U (such that U differs from S only in i’s component) C (i, S ) C (i,U ) G(S) is the equilibrium graph. 9 Example ? U C B E R K E L E Y C O M P U T E R S C I E N C E Set a=5, and consider: +1 -2 -1 -5 -1 +2 -1 +5 +5 +5 -5 +4 +1 -1 -5 -5 +1 10 Definitions: Price of Anarchy U C B E R K E L E Y C O M P U T E R S C I E N C E Price of Anarchy (Koutsoupias & Papadimitriou, 1999): the ratio between the worst-case social cost of a Nash equilibrium network and the optimum social cost over all Nash equilibria. We bound the worst-case price of anarchy to limit “the price we pay” for operating without centralized control 11 Social optima for a < 2 U C B E R K E L E Y C O M P U T E R S C I E N C E When a < 2, the social optima is a clique. Any missing edge can be added adding a to the social cost and subtracting at least 2 from social cost. A clique 12 Nash Equilibrium for 1<a<2 U C B E R K E L E Y C O M P U T E R S C I E N C E When 1<a<2, the worst-case equilibrium configuration is a star. The total cost here is (n-1)a + 2n(n-1) - 2 In a Nash Equilibrium, no single node can unilaterally add or delete an edge to bring down its cost. 13 Social optima for a > 2 U C B E R K E L E Y C O M P U T E R S C I E N C E When a > 2, the social optima is a star. Any extra edges are too expensive. C (star) (n 1)(a 2(n 1)) 14 Complexity issues U C B E R K E L E Y C O M P U T E R S C I E N C E Theorem. Computing the best response of a given peer is NP-hard. Proof hint. When 1 < a < 2, for a given node k, if there are no incoming edges, then the problem can be reduced to the Dominating Set problem. 15 Equilibria: very small a (<2) U C B E R K E L E Y C O M P U T E R S C I E N C E For a<1, the clique is the only N.E. For 1<a<2, clique no longer N.E., but the diameter is at most 2 -2 +a Then, the star is the worst N.E., can be seen to yield P.o.A. of at most 4/3 16 P.O.A for very small a (<2) U C B E R K E L E Y C O M P U T E R S C I E N C E The star is also a Nash equilibrium, but there may be worse Nash equilibrium. 17 P.O.A for very small a (<2) U C B E R K E L E Y C O M P U T E R S C I E N C E Proof. 18 The case of a > n2 U C B E R K E L E Y C O M P U T E R S C I E N C E The Nash equilibrium is a tree, and the price of anarchy is 1. Why? 19