Design and management of Noncooperative Communication Networks Ariel Orda Dept. of Electrical Engineering Technion – Israel Institute of Technology
Ariel Orda, Dept. of EE, Technion2 Queuing systems (early work) [Leeman, 1964] It is a bit surprising that in a capitalist economy, applied queuing theory limits itself to recommendations of administrative measures for the reduction of queues. One might have expected to observe such an approach in a planned economy but not in an economy in which prices and markets play so large a role. [Kleinrock, 1967] Optimum Bribing for Queue Position [Naor, 1969] Queue imposes admission fee, customers join/balk. Individual decision deviates from socially preferred one. Implicit investigation of equilibrium behavior. [Adiri & Yechiali, 1974] Optimal Priority-Purchasing and Pricing Decisions in Queues
Ariel Orda, Dept. of EE, Technion3 Transportation networks (early work) Highly individualized Each driver makes their own route choices [Wardrop, 1952] Equilibrium principle: the journey time on all routes actually used are equal, and less than those … on any unused route. System Optimum principle: The average journey time is minimum. 1 st principle analogous to Nash ’ s Equilibrium [1950]: A flow pattern is a Nash equilibrium if no individual decision maker on the network can change to a less costly route. [Beckmann, McGuire, and Winsten, 1956] Existence and uniqueness of Wardrop (Nash) equilibrium through transformation [Dafermos and Sparrow, 1969,1971] Equilibration algorithms Tolls that guarantee that the user-optimized and system-optimized solutions coincide.
Ariel Orda, Dept. of EE, Technion4 Early motivation in communication networks System-wide optimization demands coordination/cooperation among components. Impractical in large-scale networks: Large size prohibitive delays. Many components lot of information. Variability/dynamics frequent global changes. Heterogeneity of users different goals. No single administration (internetworking).
Ariel Orda, Dept. of EE, Technion5 Early motivation in communication networks (cont.) Alternative Approach: Decentralized control decisions by network components (users). Each user optimizes its performance. Research methodology: Game Theory. Network operating points: Nash equilibria.
Ariel Orda, Dept. of EE, Technion6 More recent motivation in communication networks Deregulation and privatization competitive behavior among telecom operators. Intelligence pushed to the edges of the network possibility for intelligent yet selfish decisions. Ad hoc networks: “ hosts ” are also “ nodes/routers ”. Increased interest due to the Internet boom.
Ariel Orda, Dept. of EE, Technion7 “Architecting Noncooperative Networks” Motivation: Noncooperative equilibria are inefficient and lead to suboptimal network performance. Goal: Given the noncooperative character of network control, devise design and management rules, so that the overall network performance is improved. Architect the operating points (Nash equilibria) so that they exhibit certain desirable properties. Methodology: Provisioning phase: network resource configuration. Run time phase: control part of the traffic, employ pricing mechanisms. a tility functions: just monotonic in the link flows. Parallel links: Nash equilibrium exists but is not unique. Dynamic convergence by best-reply moves from any initial profile. General-topology networks: Bottleneck-type cost functions: above results extended. Additive cost functions: Nash equilibrium need not exist.
Ariel Orda, Dept. of EE, Technion8 Motivation – the Braess Paradox The Braess Paradox [Braess, 1969]: Increasing capacity leads to performance degradation of all users. Observed in networks of various kinds: transportation networks [Dafermos & Nagurney, 1984] electrical circuits [Cohen & Horwitz, 1991] mechanical networks of strings and springs, hydraulic systems queuing networks (infinite population) [Cohen & Kelly, 1990] distributed computation systems [Glance & Hogg, 1995] telecommunication networks (finite population) [Korilis, Lazar & Orda, 1995; Cohen & Jeffries, 1997]
Ariel Orda, Dept. of EE, Technion9 The Braess Paradox (cont.)