Heuristics for Work Distribution of a Homogeneous Parallel Dynamic Programming Scheme on Heterogeneous Systems ⋆ Javier Cuenca a Domingo Gim´ enez b,1 Juan-Pedro Mart´ ınez c a Departamento de Ingenier´ ıa y Tecnolog´ ıa de Computadores, Universidad de Murcia, Spain. E-mail: javiercm@ditec.um.es b Departamento de Inform´ atica y Sistemas. Universidad de Murcia, Spain. Email: domingo@dif.um.es c Departamento de Estad´ ıstica y Matem´ atica Aplicada. Universidad Miguel Hern´ andez. Alicante, Spain. Email: jp.martinez@umh.es Abstract In this paper the possibility of including automatic optimization techniques in the design of parallel dynamic programming algorithms in heterogeneous systems is ana- lyzed. The main idea is to automatically approach the optimum values of a number of algorithmic parameters (number of processes, number of processors, processes per processor), and thus obtain low execution times. Hence, users could be pro- vided with routines which execute efficiently, and independently of the experience of the user in heterogeneous computing and dynamic programming, and which can adapt automatically to a new network of processors or a new network configuration. Key words: Auto-tuning, Heterogeneous computing, Parallel dynamic programming, Parameter optimization, Performance modelling, Scheduling, Load-balancing 1 Introduction Automatic tuning techniques have been used in the design of parallel routines in recent years. Techniques have been developed in different fields [1,2], and especially in linear algebra routines [3–5]. ⋆ Expanded version of a talk given at the HeteroPar 2004 Workshop, Cork, Ireland, July 5-8 2004. 1 Corresponding author. Tel: +34-968-36-4611; fax: +34-968-36-4151 Preprint submitted to Elsevier Science 15 March 2005