Future Generation Computer Systems 17 (2001) 415–423 A genetic-based fault-tolerant routing strategy for multiprocessor networks Peter K.K. Loh * , Venson Shaw Nanyang Technological University, School of Applied Science, Division of Computing Systems, Blk N4, 2A-36 Nanyang Avenue, Singapore 639798, Singapore Abstract AI-based search techniques have been adapted as viable, topology-independent fault-tolerant routing strategies on multipro- cessor networks [P.K.K. Loh, Artificial intelligence search techniques as fault-tolerant routing strategies, Parallel Computing 22 (8) (1996) 1127–1147]. These fault-tolerant routing strategies are viable with the exception that the routes obtained were non-minimal. This meant that a large number of redundant node traversals were made in reaching the destination, increasing the likelihood of encountering further faulty network components. Here, we investigate the adaptation of a genetic-heuristic algorithm combination as a fault-tolerant routing strategy. Our results show that this hybrid fault-tolerant routing strategy produces minimal or near-minimal routes. Under certain fault conditions, this new strategy outperforms the heuristic AI-based ones with a significant reduction in the number of redundant traversals. © 2001 Elsevier Science B.V. All rights reserved. 1. Introduction Much research has been centred on the use of genetic algorithms [3,5,16]. Few have, however, ap- plied genetic algorithms to manage routing [12–14]. In [14], a Markov model for an unreliable link is used as part of three heuristic routing table optimi- sation algorithms. Routing tables are obtained such that the overall network grade-of-service (NGOS) is minimised under different call control rules. These algorithms are used to demonstrate the improvements in trunk network fault-tolerance (assessed in terms of NGOS) that can be achieved through routing table updating in the event of link (trunk group) failure. In [13], these three algorithms are used to demonstrate the improvements in trunk network fault-tolerance that can be achieved through network augmentation, i.e. addition of extra links to a network, either in * Corresponding author. Tel.: +65-7904594; fax: +65-7926559. E-mail address: askkloh@ntu.edu.sg (P.K.K. Loh). parallel with existing links or between previously unconnected nodes. In [12], an alternative approach to routing table optimisation is taken, which makes use of a genetic algorithm. The genetic algorithm is applied indirectly as an optimisation search tech- nique to achieve dynamic routing control rather than adapted for use directly as a routing strategy. Here, the network model is assumed to be constant with no mechanism incorporated to tolerate faulty net- work components. Instead, based on traffic condi- tions at a given time, the routing control is invoked if the maximum call loss-rate (fraction of traffic that a link is unable to carry) exceeds a specified threshold. Existing research has developed several fault- tolerant routing strategies which are adaptive. How- ever, many of them are topology-dependent, i.e. the strategy is based on some regularity or geomet- rical properties inherent in the network intercon- nection structure. We have shown how a class of AI-based search techniques can be adapted for use as 0167-739X/01/$ – see front matter © 2001 Elsevier Science B.V. All rights reserved. PII:S0167-739X(99)00122-3