Simultaneous routing and resource allocation via dual decomposition AUTHOR: Lin Xiao, Student Member, IEEE, Mikael Johansson, Member, IEEE, and Stephen P. Boyd, Fellow, IEEE Reporter D 林豐裕 D 呂俊宏 IEEE Transactions on Communications, Vol. 52, No. 7, pages , July 2004

Contents Motivations System Variables System Model Design Objective and Performance Index Solution Motivation & Contribution Conclusion

Motivations Technological motivation Wireless ad-hoc networks promising emerging technology Intellectual motivation Will ad-hoc networks deliver the required performance (capacity)? Compute the optimal parameters for a given network configuration Devise simple, distributed protocols that ensure efficient network operation Control-theoretic motivation Distributed resource allocation problems roots of distributed control theory New technological challenges/problems may inspire theoretical advances Pedagogical motivation To convey ideas and techniques from distributed convex optimization

System Variables Controlled Input variables : communications variable  (or r) including media access scheme, such as transmit power, bandwidth, time slot fraction and etc., weights on flow State variables : collection of source sink vector s, collection of flow vector x, Internal variables : none Internal parameters : total node number N, total destination number D, total link number L media access methods, coding and modulation scheme, network topology A nl, Measured output variable : average behavior of data transmission, total energy consumption, total Throughput, total utilization Controlled output variables : the same variable above Design Objective : Optimization network total utility, total power consumption System: Internal Variable Input variablesOutput Variable

System model Assumption fixed topology fixed coding, modulation and optimize rates, routing & resource allocation Modeling multiple data flows influence of resource allocation on link capacities local & global resource limits

System Model- Network topology Directed graph with nodes, links set of outgoing links at node, incoming links at Incidence matrix Link, L{1,…..,m} node

System Model- Network flow model Model average data rates, multiple source/destination pairs Identify flows by destination –source flows flow from node to node –link flows flow on link to node Flow conservation laws 流量守恆定律 A is node-link incidence matrix in previous slide

System Model- Multicommodity network flow Some traditional formulations: fixed, minimize total delay: fixed, maximize total utility: Total Traffic on link l U be a concave and strictly increasing utility function Capacity and source flow k  link

System Model- Communications model Capacities determined by resource (power, bandwidth) allocation Communications model(p.1138) Where – is a vector of resources allocated to link, e.g., – is concave and increasing resource limits local (power at node) or global (total bandwidth) Many (most?) channel models satisfy these assumptions! (transmit power, bandwidth) l = 1,…..,L ; link l 的 total traffic t ≦ c l

System Model- Description Model for solving SRRA Problem Maximize weighted sum of capacities, subject to resource limits Convex problem Special methods for particular cases, e.g. water filling for variable powers, fixed bandwidth Φ are concave and monotone increasing in r w, nonnegative scalar weight

Simultaneous optimization of routing and resource allocation Solution to optimization problem We assume that are convex When communication resource allocation r is fixed, get convex multicommodity flow problem SRRA is a convex optimization problem, hence readily solved total traffic flows f(link, node, traffic)+ f(resource)

Examples SRRA formulation is very general, includes Maximum utility routing (QoS) Minimum power routing as well as minimum bandwidth, minimax link utilization, etc.

Solution - Solution Methods

Solution - Optimal Routing in a Example 50 nodes, 340 links (transmitters) 5 nodes exchange data (i.e., 20 source-destination pairs) transmitters use FDMA, power limited in each node goal: maximize network utility

Solution - Optimal Routing in a Example

Optimal Routing – Methods

Solution - Dual Decomposition Method structure of SRRA problem – objective separable in network flow and communications variables – only capacity constraints couple x; s; t and  dual decomposition (Lagrange relaxation) –relax coupling capacity constraints by introducing Lagrange multipliers( 請參考 ) –decompose SRRA into two subproblems, both highly structured, –efficient algorithms exist for each (dual decomposition again) –Sub-problems coordinated by master dual problem

Solution - Dual Decomposition Method

Solution - SRRA Solution Hierarchical Dual Decomposition Multicommodity Network Flow Resource Allocation Problem

Solution – Subproblem Method Analysis ACCPM:Analytic center cutting-plane method, Goffin and Vial [GV93] Dual Function

Conclusion Significance – New Network model considering wireless environment – Optimal operation of network using SRRA Method Problem –Do not mention the important issues in wireless network such as QoS, Dynamic Routing, media access methods, coding and modulation scheme –Cannot explain some situations – packet loss, retransmissions, time varying fading, topology change Extension – Improve algorithm – Asynchronous distributed algorithm – Dynamic routing and resource allocation

簡報結束 Thanks for Listening

Solution Introduction to Optimal Routing &resource Allocation

System Model- Example: Gaussian broadcast with FDMA Communication variables: Shannon capacity: Total power, bandwidth constraint on outgoing links (Total Resource Limits) p.1138 Shannon entropy, H1 Probability of observing a particular symbol or event, pi, with in a given sequence and

An example with FDMA

Solution - Dual Decomposition Method

Throughput, fairness, QoS End-to-end delay Gin index(fairness 指標 )