Keywords: optimization, parallel evolutionary algorithms, load-balancing, parameter identification, inverse analysis, parallel computing.

One of the possible ways to minimize the computational time for current optimization tasks is to utilize available personal computers (PCs) connected usually in a cluster. As an optimization method, evolutionary algorithms (EAs) are usually applied when dealing with objective functions which are not convex or even not continuous. There are two basic paradigms within parallel evolutionary algorithms [1]: (i) master-slave and (ii) island approaches. Since our topic deals with the identification of material parameters for very costly finite element computations, the load-balancing of a master-slave EA is presented.

The master-slave paradigm (also known as processor farming) is known within an evolutionary algorithms area for several years [2]. Also recommendations for solving the heterogeneity of a PCs cluster are already known [1]. In the case of proposed parameters identification of computer experiments we face a problem of a huge heterogeneity of the time of the objective function evaluations since the non-linear analysis time can differ with an order of magnitudes. Both heterogeneities can be efficiently solved with an asynchronous version of an evolutionary algorithm [1] where the master/root processor is not waiting for slow processors or too long computations, respectively.

Finally, different settings of synchronous and asynchronous EAs are simulated and compared. Timings data are selected from the multi-objective identification of material properties for a microplane model for concrete [3]. The results show that for a small number of available CPUs (less than five) it is more efficient to use a synchronous model than its asynchronous counterparts. With more than ten processors, more than 30% improvements in speed-up can be obtained. However, these modifications should be checked carefully since they can drastically influence the performance of the underlying optimization algorithm.

1

E. Cantú-Paz, "Efficient and Accurate Parallel Genetic Algorithms", Kluwer Academic Publishers, 2001.

2

B.H.V. Topping, A.I. Khan, "Parallel Finite Element Computations", Saxe-Coburg Publications, Edinburgh, UK, 1996.

3

M. Lepš, "Parallel Multi-Objective Identification of Material Parameters for Concrete", in B.H.V. Topping, (Editor), "Proceedings of the Ninth International Conference on the Application of Artificial Intelligence to Civil, Structural and Environmental Engineering", Civil-Comp Press, Stirlingshire, UK, Paper 33, 2007. doi:10.4203/ccp.87.33

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £72 +P&P)