Algorithmic Issues in Strategic Distributed Systems

FOURTH PART Algorithmic Issues in Strategic Distributed Systems

Suggested readings • Algorithmic Game Theory, Edited by Noam Nisan, Tim Roughgarden, Eva Tardos, and Vijay V. Vazirani, Cambridge University Press. • Blog by Noam Nisan http://agtb.wordpress.com/

Two Research Traditions • Theory of Algorithms: computational issues • What can be feasibly computed? • How long does it take to compute a solution? • Which is the quality of a computed solution? • Centralized or distributed computational models • Game Theory: interaction between self-interested individuals • What is the outcome of the interaction? • Which social goals are compatible with selfishness?

Different Assumptions • Theory of Algorithms (in distributed systems): • Processors are obedient, faulty (i.e., crash), adversarial (i.e., Byzantine), or they compete without being strategic (e.g., concurrent systems) • Large systems, limited computational resources • Game Theory: • Players are strategic(selfish) • Small systems, unlimited computational resources

The Internet World • Users often selfish • Have their own individual goals • Own network components • Internet scale • Massive systems • Limited communication/computational resources  Both strategic and computational issues!

Fundamental question • How the computational aspects of a strategic distributed system should be addressed? Algorithmic Game Theory Theory of Algorithms Game Theory = +

Basics of Game Theory • A game consists of: • A set of players (or agents) • A specification of the information available to each player • A set of rules of encounter: Who should act when, and what are the possible actions (strategies) • A specification of payoffs for each possible outcome (combination of strategies) of the game • Game Theory attempts to predict the final outcome (or solution) of the game by taking into account the individual behavior of the players

Solution concept How do we establish that an outcome is a solution? Among the possible outcomes of a game, those enjoying the following property play a fundamental role: • Equilibrium solution: strategy combination in which players are not willing to change their state. This is quite informal: when a player does not want to change his state? In the Homo Economicus model, this makes sense when he has selected a strategy that maximizes his individual payoff, knowing that other players are also doing the same.

Roadmap • We will focus on two types of equilibria: Nash Equilibria (NE) and Dominant Strategy Equilibria (DSE) • Computational Aspects of Nash Equilibria • Does a NE always exist? • Can a NE be feasibly computed, once it exists? • What about the “quality” of a NE? • Case study:Network Flow Game (i.e., selfish routing in Internet), Network Connection Games • (Algorithmic) Mechanism Design • Which social goals can be (efficiently) implemented in a strategic distributed system? • Strategy-proof mechanisms in DSE: VCG-mechanisms • Case study: Mechanism design for the Shortest Path Game

FIRST PART: (Nash) Equilibria

(Some) Types of games • Cooperative/Non-cooperative • Symmetric/Asymmetric (for 2-player games) • Zero sum/Non-zero sum • Simultaneous/Sequential • Perfect information/Imperfect information • One-shot/Repeated

Games in Normal-Form • A set of Nrational players • For each player i, a strategy set Si • Apayoff matrix: for each strategy combination (s1, s2, …, sN), where siSi, a corresponding payoff vector (p1, p2, …, pN)  |S1||S2|… |SN| payoff matrix We start by considering simultaneous, perfect-information and non-cooperative games. These games are usually represented explicitly by listing all possible strategies and corresponding payoffs of all players (this is the so-called normal–form); more formally, we have:

A famous game: the Prisoner’s Dilemma Non-cooperative, symmetric, non-zero sum, simultaneous, perfect information, one-shot, 2-player game Strategy Set Payoffs Strategy Set

Prisoner I’s decision • Prisoner I’s decision: • If II chooses Don’t Implicate then it is best to Implicate • If II chooses Implicate then it is best to Implicate • It is best to Implicate for I, regardless of what II does: Dominant Strategy

Prisoner II’s decision • Prisoner II’s decision: • If I chooses Don’t Implicatethen it is best to Implicate • If I chooses Implicatethen it is best to Implicate • It is best to Implicate for II, regardless of what I does: Dominant Strategy

Hence… • It is best for both to implicate regardless of what the other one does • Implicate is a Dominant Strategy for both • (Implicate, Implicate) becomes the Dominant Strategy Equilibrium • Note: If they might collude, then it’s beneficial for both to Not Implicate, but it’s not an equilibrium as both have incentive to deviate

Dominant Strategy Equilibrium • Dominant Strategy Equilibrium: is a strategy combination s*= (s1*, s2*, …, sN*), such that si*is a dominant strategy for each i, namely, for any possible alternativestrategy profile s= (s1, s2, …, si , …, sN): pi(s1, s2, …, si*, …, sN) ≥ pi(s1, s2, …, si, …, sN) • Dominant Strategy is the best response to any strategy of other players • If a game has a DSE, then players will immediately converge to it • Of course, not all games (only very few in the practice!) have a dominant strategy equilibrium

A more relaxed solution concept: Nash Equilibrium [1951] • Nash Equilibrium: is a strategy combination s*= (s1*, s2*, …, sN*) such that for each i, si*is a best response to (s1*, …,si-1*,si+1*,…, sN*), namely, for any possible alternativestrategy siof player i pi(s1*, s2*, …, si*, …, sN*) ≥ pi(s1*, s2*, …, si, …, sN*)

Nash Equilibrium • In a NE no agent can unilaterally deviate from his strategy given others’ strategies as fixed • Each agent has to take into consideration the strategies of the other agents • If the game is played repeatedly and players converge to a solution, then it has to be a NE • But if a game has one or more NE, players need not to converge to it • Dominant Strategy Equilibrium  Nash Equilibrium (but the converse is not true)

Nash Equilibrium: The Battle of the Sexes (coordination game) • (Stadium, Stadium) is a NE: Best responses to each other • (Cinema, Cinema) is a NE: Best responses to each other  but they are not Dominant Strategy Equilibria … are we really sure they will eventually go out together????

A crucial issue in game theory: the existence of a NE • Unfortunately, for pure strategies games (as those seen so far, in which each player, for each possible situation of the game, selects his action deterministically), it is easy to see that we cannot have a general result of existence • In other words, there may be no, one, or many NE, depending on the game

A conflictual game: Head or Tail • Player I (row) prefers to do what Player II does, while Player II prefer to do the opposite of what Player I does!  In any configuration, one of the players prefers to change his strategy, and so on and so forth…thus, there are no NE!

On the existence of a NE • However, when a player can select his strategy randomly by using a probability distribution over his set of possible pure strategies (mixed strategy), then the following general result holds: • Theorem (Nash, 1951): Any game with a finite set of players and a finite set of strategies has a NE of mixed strategies (i.e., there exists a profile of probability distributions for the players such that the expectedpayoff of each player cannot be improved by changing unilaterally the selected probability distribution). • Head or Tail game: if each player sets p(Head)=p(Tail)=1/2, then the expected payoff of each player is 0, and this is a NE, since no player can improve on this by choosing unilaterally a different randomization!

Fundamental computational issues concerned with NE • Finding a NE in mixed/pure (if any) strategies • Establishing the quality of a NE, as compared to a cooperative system, namely a system in which agents can collaborate (recall the Prisoner’s Dilemma) • In a repeated game, establishing whether and in how many steps the system will eventually converge to a NE (recall the Battle of the Sexes) • Verifying a NE, approximating a NE, NE in resource (e.g., time, space, message size) constrained settings, breaking a NE by colluding, etc... (interested in a PhD?)

Finding a NE in pure strategies • By definition, it is easy to see that an entry (p1,…,pN) of the payoff matrix is a NE if and only if pi is the maximum ith element of the row (p1,…,pi-1, {p(s):sSi} ,pi+1,…,pN), for each i=1,…,N. • Notice that, with N players, an explicit (i.e., in normal-form) representation of the payoff functions is exponential in N brute-force (i.e., enumerative) search for pure NE is then exponential in the number of players (even if it is still polynomial in the input size, but the normal-form representation needs not be a minimal-space representation of the input!)  Alternative cheaper methods are sought: for many games of interest, a NE can be found in poly-time w.r.t. to the number of players (e.g., using the powerful potential method)

On the quality of a NE • How inefficient is a NE in comparison to an idealized situation in which the players would strive to collaborate selflessly with the common goal of maximizing the social welfare? • Recall: in the Prisoner’s Dilemma (PD) game, the DSE (and NE) incurs a total of 10 years in jail for the players. However, if they would not implicate reciprocally, then they would stay a total of only 2 years in jail!

A worst-case perspective: the Price of Anarchy (PoA) • Definition (Koutsopias & Papadimitriou, 1999): Given a game G and a social-choice function C which depends on the payoff of all the players, let S be the set of all NE. If the payoff represents a cost (resp., a utility) for a player, let OPT be the outcome of G minimizing (resp., maximizing) C. Then, the Price of Anarchy (PoA) of G w.r.t. C is • Example: in the PD game, PoAPD(C)=10/2=5 PoAG(C) =

A case study for the existence and quality of a NE: selfish routing on Internet • Internet components are made up of heterogeneous nodes and links, and the network architecture is open-based and dynamic • Internet users behave selfishly: they generate traffic, and their only goal is to download/upload data as fast as possible! • But the more a link is used, the more is slower, and there is no central authority “optimizing” the data flow… • So, why does Internet eventually work is such a jungle???

Modelling the flow problem Internet can be modelled by using game theory: it is a (congestion) gamein which players users strategies paths over which users can route their traffic Non-atomic Selfish Routing: • There is a large number of (selfish) users; • All the traffic of a user is routed over a single path simultaneously; • Every user controls an infinitesimal fraction of the traffic.

Mathematical model (multicommodity flow network) • A directed graph G = (V,E) and a set of N players • A set of commodities, i.e., source–sink pairs (si,ti), for i=1,..,k (each of the N players is associated with a commodity) • An amount (or rate) 0≤ ri ≤1 of traffic between si and ti for each i=1,..,k, with i=1,…,kri = 1 • A set Pi of paths between si and ti for each i=1,..,k • The set of all paths Π=Ui=1,…,k Pi • A flow vector f specifying a traffic routing: fP: rate of traffic routed on path P (notice that 0≤ fP ≤1 ) • A flow is feasible if for everyi=1,..,k we have PPifP =ri

Mathematical model (2) • For each eE, the amount of flow absorbed by e w.r.t. f is fe=P:ePfP • For each edge e, a real-value latency function le(x) of its absorbed flow x (this is a monotonically increasing function which expresses how e gets congested when a fraction 0≤x≤1 of the total flow f uses e) • Cost (or latency) of a pathP: c(P)=eP le(fe) • Cost (or total latency) of a flowf (social-choice function): C(f)=PΠfP· c(P) = eEfe· le(fe) Observation: Noticethat the game isnotgiven in normalform!

Flows and NE Definition: A flow f* is a Nash flow if no player can improve its cost (i.e., the cost of its used path) by changing unilaterally its path. QUESTION: Given an instance(G,r=(r1,…,rk),l=(le1,…, lem)) of the non-atomic selfish routing game, does it admit a Nash flow? And in the positive case, what is the PoA of such Nash flow?

Example: Pigou’s game [1920] Latency depends on the congestion (x is the fraction of flow using the edge) le1(x)=x Total amount of flow: 1 s t Latency is fixed le2(x)=1 • What is the (only) NE of this game? Trivial: all the fraction of flow tends to travel on the upper edge  the cost of the flow is C(f) =1·le1(1)+0·le2(0)= 1·1+0·1= 1 • What is the PoA of this NE? The optimal solution is the minimum of C(x)=x·x +(1-x)·1  C’(x)=2x-1  OPT=1/2  C(OPT)=1/2·1/2+(1-1/2)·1=0.75  PoA(C) = 1/0.75 = 4/3

The Braess’s paradox • Does it help adding edges to improve the PoA? • NO!Let’s have a look at the Braess Paradox (1968) Cost of each path= x+1=1/2+1 = 1.5 v 1 x 1/2 s t 1/2 1 x Cost of the flow= 2·(1.5·1/2)=1.5 (notice this a NE and it is also an optimal flow) w

The Braess’s paradox (2) To reduce the cost of the flow, we try to add a no-latency road between v and w. Intuitively, this should not worse things! v 1 x 0 s t 1 x w

The Braess’s paradox (3) However, each user is tempted to change its route now, since the route s→v→w→thas less cost (indeed, x≤1) If only a single user changes its route, then its cost decreases from 1.5 to approximately 1, i.e.: c(s→v→w→t)= x+0+x ≈ 0.5 + 0.5 = 1 v x 1 0 s t 1 x But the problem is that all the users will decide to change! w

The Braess’s paradox (4) • So, the cost of the flow f that now entirely uses the path s→v→w→t is: C(f)=1·1+1·0+1·1=2>1.5 • Even worse, this is a NE (the cost of the path s→v→w→t is 2, and the cost of the two paths not using (v,w) is also 2)! • The optimal min-cost flow is equal to that we had before adding the new road and so, the PoA is Notice it is 4/3, as in the Pigou’s example

Existence of a Nash flow • Theorem (Beckmann et al., 1956): If for each edge e the function x·le(x) is convex (i.e., its graphic lies below the line segment joining any two points of the graphic) and continuously differentiable (i.e., its derivative exists at each point in its domain and is continuous), then the Nash flow of (G,r,l) exists and is unique, and is equal to the optimal min-cost flow of the following instance: (G,r, λ(x)=[∫ l(t)dt]/x). • Remark: The optimal min-cost flow can be computed in polynomial time through convex programming methods. x 0

Flows and Price of Anarchy • Theorem 1:In a network with linear latency functions, the cost of a Nash flow is at most 4/3 times that of the min-cost flow  every instance of the non-atomic selfish routing has PoA ≤ 4/3. • Theorem 2:In a network with degree-p polynomials latency functions, the cost of a Nash flow is O(p/log p) times that of the min-cost flow. (Roughgarden & Tardos, JACM’02)

A bad example for non-linear latencies Assume p>>1 xp 1- 1 s t 1  close to 0 0 A Nash flow(of cost 1) is arbitrarily more expensive than the optimal flow (of cost close to 0)

Convergence towards a NE(in pure strategies games) • Ok, we know that selfish routing is not so bad at its NE, but are we really sure this point of equilibrium will be eventually reached? • Convergence Time: number of moves made by the players to reach a NE from an initial arbitrary state • Question:Is the convergence time (polynomially) bounded in the number of players?

The potential function method • (Rough) Definition: A potential function for a game (if any) is a real-valued function, defined on the set of possible outcomes of the game, such that the equilibria of the game are precisely the local optima of the potential function. • Theorem: In any finite game admitting a potential function, best response dynamics (i.e., each player at each step greedily takes a move which maximizes its personal utility) always converge to a NE of pure strategies. • But how many steps are needed to reach a NE? It depends on the combinatorial structure of the players' strategy space…

Convergence towards the Nash flow Positive result: The non-atomic selfish routing game is a potential game, and moreover, for many instances (i.e., for prominent graph topologies and/or commodity specifications), the convergence time is polynomial. Negative result: However, there exist instances of the non-atomic selfish routing game for which the convergence time is exponential (under some mild assumptions).

Algorithmic Issues in Strategic Distributed Systems