Theory and Methodology An optimization problem arising in the design of multiring systems Claudio Arbib * , Fabrizio Rossi Dipartimento di Matematica Pura ed Applicata, Universit a degli Studi di L'Aquila, Via Vetoio, Coppito, I-67010 L'Aquila, Italy Received 20 July 1998; accepted 7 January 1999 Abstract We address a design problem that arises when a circular architecture is adopted to move resources in a system. In many of such architectures the demand directed from one node to another aects all the intermediate nodes, involving costs which increase both with the number of nodes and with the demand volume. We consider an application to network design and propose a model for designing a minimum cost hierarchical ring network. The problem is for- malized as that of partitioning a set of nodes into p parts to be connected by rings, with the aim of minimizing the total capacity costs. We prove that the problem is NP-complete for p P 2 and propose a guaranteed approximation algo- rithm for the case p 2. We then propose a hybrid approach employing the approximation algorithm in combination with f0; 1g LP. Experiments indicate the eectiveness of this approach in terms of computational resources and solution quality. Ó 2000 Elsevier Science B.V. All rights reserved. Keywords: Optimization; Hierarchical ring networks; Computational complexity; Approximation algorithms 1. Introduction In this paper we address a design problem that can arise when a circular architecture is adopted to move around resources (tools, parts, messages, and customers) in a system. The problem may indeed arise in various areas, such as telecom- munication systems, transportation or ¯exible manufacturing. We are here concerned with ap- plications in network design involving the use of ring networks as basic components. The architecture of a ring network connecting n nodes is such that these can be cyclically ordered, and a packet from the ith to the jth node has to go across nodes i kmod n, k 0; ... ; j i nd 1 i j. The problem of routing the packets, that is deciding such an order for each source± destination pair, is called ring loading and has been addressed in Refs. [1,2]. In the present paper we however assume that the routes agree for all source±destination pairs, i.e., that the ring is uni- directional. Such an assumption actually re¯ects a common practice of implementation according to which a (say) clockwise route is adopted for all the packets in normal conditions, whereas the anti- clockwise one is reserved for special use (e.g., failure). European Journal of Operational Research 124 (2000) 63±76 www.elsevier.com/locate/dsw * Corresponding author. Tel.: +39-0862-433-138; fax: +39- 0862-433-180. E-mail address: (arbib, rossi)@univaq.it (C. Arbib). 0377-2217/00/$ - see front matter Ó 2000 Elsevier Science B.V. All rights reserved. PII: S 0 3 7 7 - 2 2 1 7 ( 9 9 ) 0 0 1 1 9 - 8