A Constant Factor Approximation Algorithm for the Multicommodity Rent-or-Buy Problem Amit Kumar Anupam Gupta Tim Roughgarden Bell Labs CMU Cornell joint.

Slides:



Advertisements
Similar presentations
Iterative Rounding and Iterative Relaxation
Advertisements

The Primal-Dual Method: Steiner Forest TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: AA A A AA A A A AA A A.
Outline LP formulation of minimal cost flow problem
GRAPH BALANCING. Scheduling on Unrelated Machines J1 J2 J3 J4 J5 M1 M2 M3.
Introduction to Algorithms
Primal-Dual Algorithms for Connected Facility Location Chaitanya SwamyAmit Kumar Cornell University.
Multicut Lower Bounds via Network Coding Anna Blasiak Cornell University.
All-or-Nothing Multicommodity Flow Chandra Chekuri Sanjeev Khanna Bruce Shepherd Bell Labs U. Penn Bell Labs.
Approximation Some Network Design Problems With Node Costs Guy Kortsarz Rutgers University, Camden, NJ Joint work with Zeev Nutov The Open University,
1 Maximum Flow Networks Suppose G = (V, E) is a directed network. Each edge (i,j) in E has an associated ‘capacity’ u ij. Goal: Determine the maximum amount.
The Out of Kilter Algorithm in Introduction The out of kilter algorithm is an example of a primal-dual algorithm. It works on both the primal.
1 Online and Stochastic Survivable Network Design Ravishankar Krishnaswamy Carnegie Mellon University joint work with Anupam Gupta and R. Ravi.
Approximation Algorithms for Non-Uniform Buy-at-Bulk Network Design Problems Mohammad R. Salavatipour Department of Computing Science University of Alberta.
Approximation Algorithms for Non-Uniform Buy-at-Bulk Network Design Problems Guy Kortsarz Rutgers University, Camden, NJ Joint work with C. Chekuri (Bell.
Approximation Algorithm for Multicut
1 Tree Embeddings for 2-Edge-Connected Network Design Problems Ravishankar Krishnaswamy Carnegie Mellon University joint work with Anupam Gupta and R.
1 Approximation Algorithms for Demand- Robust and Stochastic Min-Cut Problems Vineet Goyal Carnegie Mellon University Based on, [Golovin, G, Ravi] (STACS’06)
Primal Dual Method Lecture 20: March 28 primaldual restricted primal restricted dual y z found x, succeed! Construct a better dual.
Network Design Adam Meyerson Carnegie-Mellon University.
14 Oct 03Cost sharing & Approximation1 Group strategy proof mechanisms via primal-dual algorithms (Cost Sharing) Martin PálÉva Tardos.
Single Sink Edge Installation Kunal Talwar UC Berkeley.
A general approximation technique for constrained forest problems Michael X. Goemans & David P. Williamson Presented by: Yonatan Elhanani & Yuval Cohen.
Robust Network Design with Exponential Scenarios By: Rohit Khandekar Guy Kortsarz Vahab Mirrokni Mohammad Salavatipour.
On the Crossing Spanning Tree Vineet Goyal Joint work with Vittorio Bilo, R. Ravi and Mohit Singh.
Maximum Flows Lecture 4: Jan 19. Network transmission Given a directed graph G A source node s A sink node t Goal: To send as much information from s.
Multicommodity Rent or Buy: Approximation Via Cost Sharing Martin Pál Joint work with Anupam Gupta Amit Kumar Tim Roughgarden.
Building Edge-Failure Resilient Networks Chandra Chekuri Bell Labs Anupam Gupta Bell Labs ! CMU Amit Kumar Cornell ! Bell Labs Seffi Naor, Danny Raz Technion.
1 On the Cost and Benefits of Procrastination: Approximation Algorithms for Stochastic Combinatorial Optimization Problems (SODA 2004) Nicole Immorlica,
Network Design with Concave Cost Functions Kamesh Munagala, Stanford University.
Primal-Dual Algorithms for Connected Facility Location Chaitanya SwamyAmit Kumar Cornell University.
A General Approach to Online Network Optimization Problems Seffi Naor Computer Science Dept. Technion Haifa, Israel Joint work: Noga Alon, Yossi Azar,
Packing Element-Disjoint Steiner Trees Mohammad R. Salavatipour Department of Computing Science University of Alberta Joint with Joseph Cheriyan Department.
Approximation Algorithms for Non-Uniform Buy-at-Bulk Network Design Problems MohammadTaghi Hajiaghayi Carnegie Mellon University Joint work with Chandra.
Integrated Logistics PROBE Princeton University, 10/31-11/1.
1 Lecture 4 Maximal Flow Problems Set Covering Problems.
Approximation Algorithms for Buy-at-Bulk Network Design MohammadTaghi Hajiaghayi Labs- Research Labs- Research.
Network Design and Bidimensionality Mohammad T. Hajiaghayi University of Maryland.
Approximation Algorithms for Stochastic Combinatorial Optimization R. Ravi Carnegie Mellon University Joint work with: Kedar Dhamdhere, CMU Kedar Dhamdhere,
ICALP'05Stochastic Steiner without a Root1 Stochastic Steiner Trees without a Root Martin Pál Joint work with Anupam Gupta.
V. V. Vazirani. Approximation Algorithms Chapters 3 & 22
LP-Based Algorithms for Capacitated Facility Location Hyung-Chan An EPFL July 29, 2013 Joint work with Mohit Singh and Ola Svensson.
1.3 Modeling with exponentially many constr.  Some strong formulations (or even formulation itself) may involve exponentially many constraints (cutting.
Network Design for Information Networks Chaitanya Swamy Caltech and U. Waterloo Ara HayrapetyanÉva Tardos Cornell University.
Approximation Algorithms Department of Mathematics and Computer Science Drexel University.
Primal-Dual Algorithms for Connected Facility Location Chaitanya SwamyAmit Kumar Cornell University.
Network Design Games Éva Tardos Cornell University.
LP-Based Algorithms for Capacitated Facility Location Hyung-Chan An EPFL July 29, 2013 Joint work with Mohit Singh and Ola Svensson.
Approximation Algorithms: Éva Tardos Cornell University problems, techniques, and their use in game theory.
Stochastic Multicast with Network Coding Ajay Gopinathan, Zongpeng Li Department of Computer Science University of Calgary ICDCS 2009, June , Montreal.
Approximation Algorithms for Prize-Collecting Forest Problems with Submodular Penalty Functions Chaitanya Swamy University of Waterloo Joint work with.
2) Combinatorial Algorithms for Traditional Market Models Vijay V. Vazirani.
1 Mar 05Stochastic Steiner without a Root1 Stochastic Steiner Tree without a Root Martin Pál Joint work with Anupam Gupta.
Multicommodity flow, well-linked terminals and routing problems Chandra Chekuri Lucent Bell Labs Joint work with Sanjeev Khanna and Bruce Shepherd Mostly.
Improved Approximation Algorithms for Directed Steiner Forest Moran Feldman Technion Joint work with: Guy Kortsarz,Rutgers University Camden Zeev Nutov,The.
12 Sep 03Martin Pál: Cost sharing & Approx1 Cost Sharing and Approximation Martin Pál joint work with Éva Tardos A-exam.
Approximating Buy-at-Bulk and Shallow-Light k-Steiner Trees Mohammad T. Hajiaghayi (CMU) Guy Kortsarz (Rutgers) Mohammad R. Salavatipour (U. Alberta) Presented.
Iterative Rounding in Graph Connectivity Problems Kamal Jain ex- Georgia Techie Microsoft Research Some slides borrowed from Lap Chi Lau.
PRIMAL-DUAL APPROXIMATION ALGORITHMS FOR METRIC FACILITY LOCATION AND K-MEDIAN PROBLEMS K. Jain V. Vazirani Journal of the ACM, 2001.
Approximation Algorithms Duality My T. UF.
Approximation Algorithms based on linear programming.
Multiroute Flows & Node-weighted Network Design Chandra Chekuri Univ of Illinois, Urbana-Champaign Joint work with Alina Ene and Ali Vakilian.
::Network Optimization:: Minimum Spanning Trees and Clustering Taufik Djatna, Dr.Eng. 1.
Theory of Computational Complexity Probability and Computing Chapter Hikaru Inada Iwama and Ito lab M1.
The minimum cost flow problem
Chapter 5. Optimal Matchings
Algorithms for Routing Node-Disjoint Paths in Grids
1.3 Modeling with exponentially many constr.
EMIS 8373: Integer Programming
Analysis of Algorithms
Introduction Basic formulations Applications
Presentation transcript:

A Constant Factor Approximation Algorithm for the Multicommodity Rent-or-Buy Problem Amit Kumar Anupam Gupta Tim Roughgarden Bell Labs CMU Cornell joint work with

The Rent-or-Buy Problem Given source-sink pairs in a network. Each pair has an associated demand. Want to install bandwidth in the network so that the source-sink pairs can communicate. t 1 t 2 s 1 s 2

The Cost Model Can rent an edge at unit cost per unit length - costs x units per unit length to send x unit of flow. Can buy an edge for M units per unit length - costs M units per unit length to send any amount of flow. t 1 t 2 s 1 s

Example t 1 t 2 s 1 s t 1 t 2 s 1 s M = 1.5 Cost = 6 Cost = 5.5

Problem Definition t 1 t 2 s 1 s Given a graph G = (V,E) and a set of source-sink pairs {(s,t )}. Reserve enough bandwidth in the network so that s can send one unit of flow to t. Objective : Minimize the total cost incurred. i i i i

Economies of Scale Edge cost a concave function of the amount of flow sent. Rent-or-buy : a non-trivial special case of this problem. has applications to multi-commodity connected facility location problem, multi-commodity maybe-cast problem [KM ’00]. bandwidth cost bandwidth cost

Special Cases t 1 t 2 s 1 s M = 1 : Always buy edges, equivalent to generalized Steiner tree problem (NP-hard). M = infinity : rent shortest source-sink paths. Our result : first constant factor approximation algorithm for the rent-or-buy problem.

Single Source Case s Edges on which bandwidth reserved form a tree. Monotonicity Property : the rented edges on any source sink path form a connected subpath. M = 2

Single Source case s M = 2 First constant factor approx. algorithm : Karger and Minkoff FOCS 2000 Constant factor improved to : Gupta,Kleinberg,Kumar,Rastogi,Yener STOC : Swamy and Kumar APPROX : Gupta, Kumar and Roughgarden STOC 2003

Rent-or-Buy Problem s 1 s 2 s 3 t 1 t 2 t 3 What properties of single source case carry over to this problem ? The set of edges used need not be a tree in any optimal solution. The set of edges that an optimal solution buys induce a forest.

Monotonicity ? The set of bought edges in any source to sink path forms a connected subpath. The definition of monotonicity must be symmetric with respect to sources and sinks.

Monotonicity t 3 t 2 t 1 s 3 s 2 s 1 M = 1.5 Theorem : There is a monotone solution whose cost is at most twice the cost of the optimal solution. Optimal solution may not be monotone.

Monotone Solution Look for optimal monotone solution only. Enough to know the component of bought edges. We can think of the vertices in these components as facilities.

Facility Location Problem Want to find facility nodes F. Each source or sink attaches to a node in F and pays the shortest distance to this node. For any (s,t) pair, the facilities to which s and t attach must be connected by bought edges – pay M times the length of edges bought.

Integer Programming Formulation Decision variables with values either 0 or 1. x : 1 if demand j assigned to i. z : 1 if the edge e is bought. ij e Objective function :

Constraints s i S t k k

Related Work The LP follows along the line of the LP given for the single source special case by [GKKRY 01]. They round the fractional solution with a constant factor loss. This technique does not extend to our problem. We use primal-dual method to show that the integrality gap of the LP is a constant. The primal-dual algorithm for the single source case given by [SK02] -- our algorithm derives intuition from this though requires several non-trivial ideas.

Dual for the rent-or-buy problem

For each demand j we associate a set S(j). If  c, then S(j) contains i, S(j) may contain other nodes. When we raise  we also raise y at the same rate. the set S(j) grows as  grows. Dual variables S j ij S(j) j j j s k t k The sets S(s ) and S(t ) are always disjoint. k k

Growing dual variables We raise  in a continuous way. Can think of S(j) as a sub-graph which grows in a continuous way. Can always choose to add new vertices to S(j). j

Growing dual variables We raise  in a continuous way. Can think of S(j) as a sub-graph which grows in a continuous way. Can always choose to add new vertices to S(j). j

Growing dual variables

Dual Feasibility M = 2 For any edge e, there will be at most M demands j such that e will ever appear in the set  (S(j)).

Primal-dual Algorithm All demands occur in groups of M. All demand pairs have the same source. Special case : Should be able to simulate Steiner tree primal-dual algorithm.

Primal-dual Algorithm

When two components merge, raise only M dual variables in it.

Primal-dual Algorithm

Problems If (s,t) is a demand pair, can raise their dual variables as long as their sub-graphs are disjoint. Even if demands occur in groups of M, this may not be the case when we raise the dual variables. Suppose demands do not have a common source. Demands may not occur in groups of M. – raise dual variables so that they form groups of size M.

Primal-dual Algorithm Raise the dual variables such that clusters of at least M demands form. Collapse the clusters into single nodes and simulate the Steiner forest primal-dual algorithm as long as each component has at least M nodes. The algorithm alternates between the two steps.

Primal-dual Algorithm M = 3 Consolidate demands into groups of M.

Primal-dual Algorithm M = 3 Collapse clusters into single nodes.

Primal-dual Algorithm M = 3 Simulate primal-dual algorithm for the Steiner forest problem.

Primal-dual Algorithm M = 3 Cancel demands.

Primal-dual Algorithm M = 3 Consolidate demands again.

Primal-dual algorithm In a given phase, many edges may become unusable because of previous phases. Need to bound the effect of previous phases on the current phase -- make sure that the dual variables increase by an exponential factor in successive phases.

Open Problems buy-at-bulk network design : have many different types of cables, each has a fixed cost and an incremental cost. Single source buy-at-bulk : constant factor approximation algorithm known [Guha Meyerson Munagala ’01]. bandwidth cost

Open Problems Buy-at-bulk : even for single cable nothing better than logarithmic factor known [Awerbuch Azar ’97]. Single source rent-or-buy : what if different edges have different buying costs ? Nothing better than logarithmic factor known [Meyerson Munagala Plotkin ’00].