Abstract— This paper presents a new parallel evolutionary approach where the concept of generation has been removed and replaced by the cycle one. Indeed, the classical genetic algorithms (GAs) deals with operations on the whole population through all generations. These operations are performed during the evolution towards the best individual or solution of the considered combinatorial problem. In our approach, each individual participates to the evolutionary process uniquely during some iterations. There is no generation where all individuals are created at the same time and disappear at the same time at the end of the evolutionary process genation. In our approach, each individual owns one lifespan represented by a number of cycles which are affected to it randomly at its birth and at the end of which it disappears from the population. Consequently, only certain individuals of the population are evaluated within each iteration of the algorithm and not all the population. This causes the substantial reduction of the total running time of the algorithm since the evaluations of all individuals of each generation necessitates more than 80% of the total running time of a classical GA. This approach has been developed with the goal to present a new and efficient parallel scheme of the classical GA with better performances in terms of running time. In this paper, we will present a new asynchronous parallel Master/Slave scheme of the GA and will show the power of our approach with the classification extraction rules problem. I. INTRODUCTION The genetic algorithm [1] [2] is a robust meta heuristic which proved its robustness in large size optimization combinatorial problems. Indeed, it is an efficient method which explores the research space effectively, while converging towards an appreciable solution. However, the major disadvantage of the GAs is sometimes its total running time which may be long for large size problems. Indeed, for each generation, it is necessary to carry out all genetic operations using all the population chromosomes, and especially computing each individual fitness value through all the generations. Generally, the fitness calculation requires a considerable time. In addition, it is necessary to evaluate all the chromosomes of a generation to be able to S. Benkhider, is with the Artificial Intelligence Research Laboratory of the Computer Sciences Department, University of Sciences and Technology Houari Boumedienne, Algiers, Algeria, e-mail:sadjiab@yahoo.com A.R. Baba-Ali, is with the LRPE Laboratory, of Electronic Institute, University of Sciences and Technology Houari Boumedienne, Algiers, Algeria, e-mail:riadhbabaali@yahoo.fr H. Drias is with the National Computer Scineces Institute, OuedSmar, Algiers, Algeria. carry out the selections of the bests in order to constitute the new population. Consequently, when processing a current generation, no genetic operations on the future generation can be started,. This is even true with the parallel version of a classical GA which consists of evaluating the chromosomes in parallel using slave machines. This latency represents a disadvantage because it should be waited until all the evaluations were returned to the master process. This is necessary in order to be able to start the selection process. In this paper, we describe a new parallel evolutionary approach with aim to minimize the total running time of classical GAs while maintaining the same accuracy. II. NEW APPROACH DESCRIPTION The approach we present, proposes some improvements of the classical genetic algorithm, with the aim of decreasing its running time, while maintaining a good quality of the found solution. The algorithm makes an evolution of individuals on several iterations rather than generations. In our approach, we define a new important parameter that is: the individual maximum lifespan. This value represents the maximum number of iterations where an individual can participate to the evolutionary process. So, each individual is randomly assigned a lifespan ranging from 1 iteration to the maximum lifespan. If the iteration count of an individual reaches its lifespan, it then "disappears" from the population and doesn’t participate to the evolutionary process anymore. Our population is dynamic in which several kinds of individuals take part in the evolutionary process, through several iterations. A. Life-Cycle of an individual The individual cycle of life is characterized by three stages: Birth Life End of Lifetime The birth means the production of new individual in the population. One lifespan and a date of completion of life are granted to it: - lifespan: The lifespan of an individual is the iteration count during which it can "live" (or exist in the population). It is a positive integer value randomly chosen in the interval [ 1, max_lifespan] where max_lifespan is the maximum iteration count that the individual can live in the population. A new generationless parallel evolutionary algorithm for combinatorial optimization S.Benkhider 1 , A.R. Baba-Ali 2 , H. Drias 1 4691 1-4244-1340-0/07$25.00 c 2007 IEEE