Designing Fast Converging Cost Sharing Methods for Multicast Transmissions

We study a multicast game in non-cooperative directed networks in which a source sends the same message or service to a set of r receiving users and the cost of the used links is divided among the receivers according to a given cost sharing method. By following the approach recently proposed by Chen et al. (Proceedings of the 19th Annual ACM-SIAM Symposium on Discrete Algorithms (SODA), pp. 854–863, 2008), we analyze the performances of a family of methods satisfying certain desiderata, namely, weak and strong budget-balance, fairness and separability. We show that any fair method may require an arbitrary number of selfish moves in order to converge to a pure Nash equilibrium, hence we focus on the solutions obtained after one round of selfish moves. We evaluate their quality according to two global social functions: the overall cost of the solution and the maximum shared cost of users. The only method satisfying all the properties is the well-known Shapley value for which we show an approximation ratio of the solutions reached after a one round walk equal to Θ(r ²). We then prove that relaxing the strong budget balance and separability properties (we call feasible any method satisfying weak budget balance and fairness) leads to improved performances since we determine a feasible method achieving an approximation ratio of the solutions reached after a one round walk equal to O(r). This bound is asymptotically optimal since we also show that any method satisfying weak budget balance cannot achieve an approximation ratio of the solutions reached after a one round walk smaller than r. Finally, we prove the NP-hardness of computing the sequence of moves leading to the best possible global performance and extend most of the results to undirected networks.     

Competitive Cost Sharing with Economies of Scale

We consider a general class of non-cooperative buy-at-bulk cost sharing games, in which k players make investments to purchase a set of resources. Each resource has a certain cost and must bought to be available to the players. Each player has a certain constraint on the number and types of resources that she needs to have available, and she can specify payments to make a resource available to her. She strives to fulfill her constraint with the smallest investment possible. Our model includes a natural economy of scale: for a subset of players capacity must be installed at the resources, and the cost increase for a resource r is composed of a fixed price c(r) and a global concave capacity function g. This cost can be shared arbitrarily between players.     

Stackelberg Strategies for Selfish Routing in General Multicommodity Networks

A natural generalization of the selfish routing setting arises when some of the users obey a central coordinating authority, while the rest act selfishly. Such behavior can be modeled by dividing the users into an α fraction that are routed according to the central coordinator’s routing strategy (Stackelberg strategy), and the remaining 1−α that determine their strategy selfishly, given the routing of the coordinated users. One seeks to quantify the resulting price of anarchy, i.e., the ratio of the cost of the worst traffic equilibrium to the system optimum, as a function of α. It is well-known that for α=0 and linear latency functions the price of anarchy is at most 4/3 (J. ACM 49, 236–259, 2002). If α tends to 1, the price of anarchy should also tend to 1 for any reasonable algorithm used by the coordinator. We analyze two such algorithms for Stackelberg routing, LLF and SCALE. For general topology networks, multicommodity users, and linear latency functions, we show a price of anarchy bound for SCALE which decreases from 4/3 to 1 as α increases from 0 to 1, and depends only on α. Up to this work, such a tradeoff was known only for the case of two nodes connected with parallel links (SIAM J. Comput. 33, 332–350, 2004), while for general networks it was not clear whether such a result could be achieved, even in the single-commodity case. We show a weaker bound for LLF and also some extensions to general latency functions. The existence of a central coordinator is a rather strong requirement for a network. We show that we can do away with such a coordinator, as long as we are allowed to impose taxes (tolls) on the edges in order to steer the selfish users towards an improved system cost. As long as there is at least a fraction α of users that pay their taxes, we show the existence of taxes that lead to the simulation of SCALE by the tax-payers. The extension of the results mentioned above quantifies the improvement on the system cost as the number of tax-evaders decreases. Research of G. Karakostas supported by an NSERC Discovery Grant and MITACS. Research of S. Kolliopoulos partially supported by the University of Athens under the project Kapodistrias.     

Stackelberg Strategies and Collusion in Network Games with Splittable Flow

We study the impact of collusion in network games with splittable flow and focus on the well established price of anarchy as a measure of this impact. We first investigate symmetric load balancing games and show that the price of anarchy is at most m, where m denotes the number of coalitions. For general networks, we present an instance showing that the price of anarchy is unbounded, even in the case of two coalitions. If latencies are restricted to polynomials with nonnegative coefficients and bounded degree, we prove upper bounds on the price of anarchy for general networks, which improve upon the current best ones except for affine latencies.     

Distributed MST for constant diameter graphs

This paper considers the problem of distributively constructing a minimum-weight spanning tree (MST) for graphs of constant diameter in the bounded-messages model, where each message can contain at most B bits for some parameter B. It is shown that the number of communication rounds necessary to compute an MST for graphs of diameter 4 or 3 can be as high as and , respectively. The asymptotic lower bounds hold for randomized algorithms as well. On the other hand, we observe that O(log n) communication rounds always suffice to compute an MST deterministically for graphs with diameter 2, when B = O(log n). These results complement a previously known lower bound of for graphs of diameter Ω(log n). An extended abstract of this work appears in Proceedings of 20th ACM Symposium on Principles of Distributed Computing, August 2001.     

On Equilibria for ADM Minimization Games

In the ADM minimization problem the input is a set of arcs along a directed ring. The input arcs need to be partitioned into non-overlapping chains and cycles so as to minimize the total number of endpoints, where a k-arc cycle contributes k endpoints and a k-arc chain contains k+1 endpoints. We study ADM minimization problem both as non-cooperative and cooperative games. In these games each arc corresponds to a player, and the players share the cost of the ADM switches. We consider two cost allocation models, a model which was considered by Flammini et al., and a new cost allocation model, which is inspired by congestion games. We compare the price of anarchy and price of stability in the two cost allocation models, as well as the strong price of anarchy and the strong price of stability.     

Optimality in External Memory Hashing

Hash tables on external memory are commonly used for indexing in database management systems. In this paper we present an algorithm that, in an asymptotic sense, achieves the best possible I/O and space complexities. Let B denote the number of records that fit in a block, and let N denote the total number of records. Our hash table uses I/Os, expected, for looking up a record (no matter if it is present or not). To insert, delete or change a record that has just been looked up requires I/Os, amortized expected, including I/Os for reorganizing the hash table when the size of the database changes. The expected external space usage is times the optimum of N/B blocks, and just O(1) blocks of internal memory are needed.     

The Network as a Storage Device: Dynamic Routing with Bounded Buffers

We study dynamic routing in store-and-forward packet networks where each network link has bounded buffer capacity for receiving incoming packets and is capable of transmitting a fixed number of packets per unit of time. At any moment in time, packets are injected at various network nodes with each packet specifying its destination node. The goal is to maximize the throughput, defined as the number of packets delivered to their destinations. In this paper, we make some progress on throughput maximization in various network topologies. Let n and m denote the number of nodes and links in the network, respectively. For line networks, we show that Nearest-to-Go (NTG), a natural distributed greedy algorithm, is -competitive, essentially matching a known lower bound on the performance of any greedy algorithm. We also show that if we allow the online routing algorithm to make centralized decisions, there is a randomized polylog(n)-competitive algorithm for line networks as well as for rooted tree networks, where each packet is destined for the root of the tree. For grid graphs, we show that NTG has a competitive ratio of while no greedy algorithm can achieve a ratio better than . Finally, for arbitrary network topologies, we show that NTG is -competitive, improving upon an earlier bound of O(mn). An extended abstract appeared in the Proceedings of the 8th Workshop on Approximation Algorithms for Combinatorial Optimization Problems, APPROX 2005, Berkeley, CA, USA, pp. 1–13, Lecture Notes in Computer Science, vol. 1741, Springer, Berlin. S. Angelov is supported in part by NSF Career Award CCR-0093117, NSF Award ITR 0205456 and NIGMS Award 1-P20-GM-6912-1. S. Khanna is supported in part by an NSF Career Award CCR-0093117, NSF Award CCF-0429836, and a US-Israel Binational Science Foundation Grant. K. Kunal is supported in part by an NSF Career Award CCR-0093117 and NSF Award CCF-0429836.     

When ignorance helps: Graphical multicast cost sharing games

In non-cooperative games played on highly decentralized networks the assumption that each player knows the strategy adopted by any other player may be too optimistic or even infeasible. In such situations, the set of players of which each player knows the chosen strategy can be modeled by means of a social knowledge graph in which nodes represent players and there is an edge from i to j if i knows j. Following the framework introduced in [7], we study the impact of social knowledge graphs on the fundamental multicast cost sharing game in which all the players want to receive the same communication from a given source in an undirected network. In the classical complete information case, such a game is known to be highly inefficient, since its price of anarchy can be as high as the total number of players ρ. We first show that, under our incomplete information setting, pure Nash equilibria always exist only if the social knowledge graph is directed acyclic (DAG). We then prove that the price of stability of any DAG is at least and provide a DAG lowering the classical price of anarchy to a value between and log²ρ. If specific instances of the game are concerned, that is if the social knowledge graph can be selected as a function of the instance, we show that the price of stability is at least , and that the same bound holds also for the price of anarchy of any social knowledge graph (not only DAGs). Moreover, we provide a nearly matching upper bound by proving that, for any fixed instance, there always exists a DAG yielding a price of anarchy less than 4. Our results open a new window on how the performances of non-cooperative systems may benefit from the lack of total knowledge among players.     

The Price of Anarchy in Network Creation Games Is (Mostly) Constant

We study the price of anarchy and the structure of equilibria in network creation games. A network creation game is played by n players {1,2,…,n}, each identified with a vertex of a graph (network), where the strategy of player i, i=1,…,n, is to build some edges adjacent to i. The cost of building an edge is α>0, a fixed parameter of the game. The goal of every player is to minimize its creation cost plus its usage cost. The creation cost of player i is α times the number of built edges. In the SumGame variant, the usage cost of player i is the sum of distances from i to every node of the resulting graph. In the MaxGame variant, the usage cost is the eccentricity of i in the resulting graph of the game. In this paper we improve previously known bounds on the price of anarchy of the game (of both variants) for various ranges of α, and give new insights into the structure of equilibria for various values of α. The two main results of the paper show that for α>273?n all equilibria in SumGame are trees and thus the price of anarchy is constant, and that for α>129 all equilibria in MaxGame are trees and the price of anarchy is constant. For SumGame this answers (almost completely) one of the fundamental open problems in the field—is price of anarchy of the network creation game constant for all values of α?—in an affirmative way, up to a tiny range of α.     

On the Performance of Approximate Equilibria in Congestion Games

We study the performance of approximate Nash equilibria for congestion games with polynomial latency functions. We consider how much the price of anarchy worsens and how much the price of stability improves as a function of the approximation factor ε. We give tight bounds for the price of anarchy of atomic and non-atomic congestion games and for the price of stability of non-atomic congestion games. For the price of stability of atomic congestion games we give non-tight bounds for linear latencies. Our results not only encompass and generalize the existing results of exact equilibria to ε-Nash equilibria, but they also provide a unified approach which reveals the common threads of the atomic and non-atomic price of anarchy results. By expanding the spectrum, we also cast the existing results in a new light.     

Selfish Load Balancing and Atomic Congestion Games

We revisit a classical load balancing problem in the modern context of decentralized systems and self-interested clients. In particular, there is a set of clients, each of whom must choose a server from a permissible set. Each client has a unit-length job and selfishly wants to minimize its own latency (job completion time). A server's latency is inversely proportional to its speed, but it grows linearly with (or, more generally, as the pth power of) the number of clients matched to it. This interaction is naturally modeled as an atomic congestion game, which we call selfish load balancing. We analyze the Nash equilibria of this game and prove nearly tight bounds on the price of anarchy (worst-case ratio between a Nash solution and the social optimum). In particular, for linear latency functions, we show that if the server speeds are relatively bounded and the number of clients is large compared with the number of servers, then every Nash assignment approaches social optimum. Without any assumptions on the number of clients, servers, and server speeds, the price of anarchy is at most 2.5. If all servers have the same speed, then the price of anarchy further improves to We also exhibit a lower bound of 2.01. Our proof techniques can also be adapted for the coordinated load balancing problem under L₂ norm, where it slightly improves the best previously known upper bound on the competitive ratio of a simple greedy scheme.     

On the complexity of distributed stable matching with small messages

We consider the distributed complexity of the stable matching problem (a.k.a. “stable marriage”). In this problem, the communication graph is undirected and bipartite, and each node ranks its neighbors. Given a matching of the nodes, a pair of unmatched nodes is called blocking if they prefer each other to their assigned match. A matching is called stable if it does not induce any blocking pair. In the distributed model, nodes exchange messages in each round over the communication links, until they find a stable matching. We show that if messages may contain at most B bits each, then any distributed algorithm that solves the stable matching problem requires ${\Omega(\sqrt{n/B\log n})}We consider the distributed complexity of the stable matching problem (a.k.a. “stable marriage”). In this problem, the communication graph is undirected and bipartite, and each node ranks its neighbors. Given a matching of the nodes, a pair of unmatched nodes is called blocking if they prefer each other to their assigned match. A matching is called stable if it does not induce any blocking pair. In the distributed model, nodes exchange messages in each round over the communication links, until they find a stable matching. We show that if messages may contain at most B bits each, then any distributed algorithm that solves the stable matching problem requires W(?{n/Blogn}){\Omega(\sqrt{n/B\log n})} communication rounds in the worst case, even for graphs of diameter O(log n), where n is the number of nodes in the graph. Furthermore, the lower bound holds even if we allow the output to contain O(?n){O(\sqrt n)} blocking pairs, and if a pair is considered blocking only if they like each other much more then their assigned match.     

Remotely Prepared Entanglement: a Quantum Web Page

Abstract. In quantum teleportation, an unknown quantum state is transmitted from one party to another using only local operations and classical communication, at the cost of shared entanglement. Is it possible similarly, using an N party entangled state, to have the state retrievable by any of the N-1 possible receivers? If the receivers cooperate, and share a suitable state, this can be done reliably. The N party GHZ is one such state; I derive a large class of such states, and show that they are in general not equivalent to the GHZ. I also briefly discuss the problem where the parties do not cooperate, and the relationship to multipartite entanglement quantification. I define a new set of entanglement monotones, the entanglements of preparation.     

Coordination mechanisms for selfish scheduling

In machine scheduling, a set of jobs must be scheduled on a set of machines so as to minimize some global objective function, such as the makespan, which we consider in this paper. In practice, jobs are often controlled by independent, selfishly acting agents, which each select a machine for processing that minimizes the (expected) completion time. This scenario can be formalized as a game in which the players are job owners, the strategies are machines, and a player’s disutility is the completion time of its jobs in the corresponding schedule. The equilibria of these games may result in larger-than-optimal overall makespan. The price of anarchy is the ratio of the worst-case equilibrium makespan to the optimal makespan. In this paper, we design and analyze scheduling policies, or coordination mechanisms, for machines which aim to minimize the price of anarchy of the corresponding game. We study coordination mechanisms for four classes of multiprocessor machine scheduling problems and derive upper and lower bounds on the price of anarchy of these mechanisms. For several of the proposed mechanisms, we also prove that the system converges to a pure-strategy Nash equilibrium in a linear number of rounds. Finally, we note that our results are applicable to several practical problems arising in communication networks.     

A Note on Multiflows and Treewidth

We consider multicommodity flow problems in capacitated graphs where the treewidth of the underlying graph is bounded by r. The parameter r is allowed to be a function of the input size. An instance of the problem consists of a capacitated graph and a collection of terminal pairs. Each terminal pair has a non-negative demand that is to be routed between the nodes in the pair. A class of optimization problems is obtained when the goal is to route a maximum number of the pairs in the graph subject to the capacity constraints on the edges. Depending on whether routings are fractional, integral or unsplittable, three different versions are obtained; these are commonly referred to respectively as maximum MCF, EDP (the demands are further constrained to be one) and UFP. We obtain the following results in such graphs.

Competitive Online Multicommodity Routing

We study online multicommodity routing problems in networks, in which commodities have to be routed sequentially. The flow of each commodity can be split on several paths. Arcs are equipped with load dependent price functions defining routing costs, which have to be minimized. We discuss a greedy online algorithm that routes each commodity by minimizing a convex cost function that depends on the previously routed flow. We present a competitive analysis of this algorithm showing that for affine price functions this algorithm is  -competitive, where K is the number of commodities. For networks with two nodes and parallel arcs, this algorithm is optimal. Without restrictions on the price functions and network, no algorithm is competitive. We then investigate a variant in which the demands have to be routed unsplittably. In this case, it is NP-hard to compute the offline optimum. The variant of the greedy algorithm that produces unsplittable flows is -competitive, and we prove a lower bound of 2 for the competitive ratio of any deterministic online algorithm. A preliminary version of this paper appeared in [20]. S. Heinz has been supported by the DFG Research Center Matheon Mathematics for key technologies in Berlin.     

Approximation Algorithms for <Emphasis Type="Italic">k</Emphasis>-hurdle Problems

The polynomial-time solvable k-hurdle problem is a natural generalization of the classical s-t minimum cut problem where we must select a minimum-cost subset S of the edges of a graph such that |p∩S|≥k for every s-t path p. In this paper, we describe a set of approximation algorithms for “k-hurdle” variants of the NP-hard multiway cut and multicut problems. For the k-hurdle multiway cut problem with r terminals, we give two results, the first being a pseudo-approximation algorithm that outputs a (k−1)-hurdle solution whose cost is at most that of an optimal solution for k hurdles. Secondly, we provide a 2(1-\frac1r)2(1-\frac{1}{r})-approximation algorithm based on rounding the solution of a linear program, for which we give a simple randomized half-integrality proof that works for both edge and vertex k-hurdle multiway cuts that generalizes the half-integrality results of Garg et al. for the vertex multiway cut problem. We also describe an approximation-preserving reduction from vertex cover as evidence that it may be difficult to achieve a better approximation ratio than 2(1-\frac1r)2(1-\frac{1}{r}). For the k-hurdle multicut problem in an n-vertex graph, we provide an algorithm that, for any constant ε>0, outputs a ⌈(1−ε)k⌉-hurdle solution of cost at most O(log n) times that of an optimal k-hurdle solution, and we obtain a 2-approximation algorithm for trees.     

Non-clairvoyant Scheduling Games

In a scheduling game, each player owns a job and chooses a machine to execute it. While the social cost is the maximal load over all machines (makespan), the cost (disutility) of each player is the completion time of its own job. In the game, players may follow selfish strategies to optimize their cost and therefore their behaviors do not necessarily lead the game to an equilibrium. Even in the case there is an equilibrium, its makespan might be much larger than the social optimum, and this inefficiency is measured by the price of anarchy—the worst ratio between the makespan of an equilibrium and the optimum. Coordination mechanisms aim to reduce the price of anarchy by designing scheduling policies that specify how jobs assigned to a same machine are to be scheduled. Typically these policies define the schedule according to the processing times as announced by the jobs. One could wonder if there are policies that do not require this knowledge, and still provide a good price of anarchy. This would make the processing times be private information and avoid the problem of truthfulness. In this paper we study these so-called non-clairvoyant policies. In particular, we study the RANDOM policy that schedules the jobs in a random order without preemption, and the EQUI policy that schedules the jobs in parallel using time-multiplexing, assigning each job an equal fraction of CPU time.     

Assignment Games with Conflicts: Robust Price of Anarchy and Convergence Results via Semi-Smoothness

We study assignment games in which jobs select machines, and in which certain pairs of jobs may conflict, which is to say they may incur an additional cost when they are both assigned to the same machine, beyond that associated with the increase in load. Questions regarding such interactions apply beyond allocating jobs to machines: when people in a social network choose to align themselves with a group or party, they typically do so based upon not only the inherent quality of that group, but also who amongst their friends (or enemies) chooses that group as well. We show how semi-smoothness, a recently introduced generalization of smoothness, is necessary to find tight bounds on the robust price of anarchy, and thus on the quality of correlated and Nash equilibria, for several natural job-assignment games with interacting jobs. For most cases, our bounds on the robust price of anarchy are either exactly 2 or approach 2. We also prove new convergence results implied by semi-smoothness for our games. Finally we consider coalitional deviations, and prove results about the existence and quality of strong equilibrium.