High-performance adaptive routing for networks with arbitrary topology V. Puente a , J.A. Gregorio a , F. Vallejo a , R. Beivide a , C. Izu b, * a Computer Architecture Group, University of Cantabria, 39005 Santander, Spain b Computer Science Department, University of Adelaide, SA 50005, Australia Available online 14 November 2005 Abstract A strategy to implement adaptive routing in irregular networks is presented and analyzed in this work. A simple and widely applicable deadlock avoidance method, applied to a ring embedded in the network topology, constitutes the basis of this high-performance packet switching. This adaptive router improves the network capabilities by allocating more resources to the fastest and most used virtual network, thus narrowing the performance gap with regular topologies. A thorough simulation process, which obtains statistically reliable measurements of irregular network behavior, has been car- ried out to evaluate it and compare with other state-of-the-art techniques. In all the experiments, our router exhibited the best behavior in terms of maximum/sustained performance and sensitivity to the network topology. Ó 2005 Elsevier B.V. All rights reserved. Keywords: Irregular topologies; Routing, Flow control; Deadlock; Cluster computing; Networks of workstations; Local area networks 1. Introduction Networks of workstations and other forms of cluster computing are currently emerging in the high-performance computer market as good alter- natives to sophisticated parallel computers. The cost/performance ratio of the commodity hardware and the existence of affordable and scalable high- performance communication technologies justify the penetration in the market of these distributed computing platforms. Moreover, their versatility makes this kind of systems particularly attractive to solve a wide range of applications. Computer clusters are commonly organized as switched networks in which each switch or router has several computing nodes connected to some of its input/output ports. The remaining ports are used to link other routers in order to provide a connected system. Bi-directional full-duplex links are com- monly employed to exploit communication locality. Packets interchanged among computing nodes cross the network according to the rules dictated by a cer- tain routing mechanism. It is well known that the interconnection network architecture and its associ- ated software libraries are critical components for high-performance cluster computing. 1383-7621/$ - see front matter Ó 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.sysarc.2005.09.003 * Corresponding author. Fax: +34 942 201479. E-mail addresses: vpuente@atc.unican.es (V. Puente), jagm@ atc.unican.es (J.A. Gregorio), fernando@atc.unican.es (F. Val- lejo), mon@atc.unican.es (R. Beivide), cruz@cs.adelaide.edu.au (C. Izu). Journal of Systems Architecture 52 (2006) 345–358 www.elsevier.com/locate/sysarc