Coordinated aggressive bidding in distributed combinatorial resource allocation Jinbo Chen, Alejandro Bugacov, Pedro Szekely, Martin Frank, Min Cai, Donghan Kim, Robert Neches USC/Information Sciences Institute 4676 Admiralty Way Marina del Rey, CA 90292 {jinbo,bugacov,szekely,frank,mcai,donghank,rneches}@isi.edu ABSTRACT Large NP-hard combinatorial resource allocation problems are best solved via approximation techniques which pro- duce acceptable solutions in the time available. The best of such “good-enough/soon-enough” techniques handle large problems, run in distributed environments, adapt rapidly to changes to the problem while solving, and exhibit good anytime performance. The Dynamic Marble Size (DMS) algorithm appears promis- ing as an approach offering all those traits. It is a market- inspired distributed multi-agent scheme, in which task agents aggressively bid for their preferred resource bundles through single-resource auctions, coordinating their interdependent bids by bid adjustment. In the DMS algorithm, tasks maximally bid their value. This aggressive bidding strategy maximizes other bidders’ information about their prospects of succeeding, and thereby also help them focus on resources they can win. An oscillation- avoiding bid adjustment algorithm utilizes a binary search technique to prevent those adjustments from introducing cycles or deadlocks. A “stubbornness-detection monitor” (a re-start limit) limits how many sets of resources a task will pursue. Aggressive bidding, oscillation-avoidance and stubbornness-detection promote rapid convergence on good solutions – bidders avoid highly-contended resources and fo- cus upon more promising alternatives. To evaluate the DMS algorithm, we analyzed characteris- tics of randomly-generated problems, using results obtained with a Pseudo-Boolean encoding of the problem as the gold standard for quality obtainable from centralized solutions. For the types of problems that our problem generator can produce, DMS produces comparable solution quality, using significantly less time, as well as exhibiting a good anytime performance. Furthermore, a variant of DMS scales linearly in the problem size. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Copyright 2002 ACM X-XXXXX-XX-X/XX/XX ...$5.00. Categories and Subject Descriptors C.2.4 [Computer Systems Organization]: Distributed Systems—distributed applications ; D.2.8 [Software Engi- neering]: Metrics—complexity measures,performance mea- sures ; I.2.11 [Computing Methodologies]: Distributed Artificial Intelligence—coherence and coordination,multiagent systems General Terms Cooperative negotiation, distributed resource allocation, mar- ket mechanisms for multi-agent systems, combinatorial auc- tions Keywords decentralized resource allocation, auction protocols, bidding strategies, distributed constraint optimization 1. INTRODUCTION A resource allocation problem consists of a set of tasks to be performed and a set of resources that can be used to perform the tasks. The goal is to determine an assign- ment of resources to tasks so that the utility of the tasks is maximized. In many real-world domains (e.g. flight crew training scheduling [2], electricity markets [4], transporta- tion exchanges [8]), resource utilization efficiency could be enhanced if tasks are allowed to express preferences over bundles of resource items(i.e, some resources can be com- plementary or substitutes). Because of resource complementarities, bids from a task are not independent. For example, a task has a bundle re- source preference: A, B and C. So bidding on a resource (B) could depend on the following: 1. whether the task is winning on resource A; 2. how much the task has bid on resource A; 3. what is the task’s speculation or counter-speculation on other tasks’ bidding on A and C. These complications cause the traditional single-resource auc- tion to be inefficient for the resource allocation problems with bundled resource preferences. To address inefficiencies of the traditional single-resource auction, combinatorial auctions were proposed, in which