Abstract
The problem of finding a global minimum of a real function on a set S Rn occurs in many real world problems. Since its computational complexity is exponential, its solution can be a very expensive computational task. In this paper, we introduce a parallel algorithm that exploits the latest computers in the market equipped with more than one processor, and used in clusters of computers. The algorithm belongs to the improvement of local minima algorithm family, and carries on local minimum searches iteratively but trying not to find an already found local optimizer. Numerical experiments have been carried out on two computers equipped with four and six processors; fourteen configurations of the computing resources have been investigated. To evaluate the algorithm performances the speedup and the efficiency are reported for each configuration.
Highlights
In this paper we consider the following global optimization problem
Two desktop computers have been used; the first equipped with an Intel Quad CPU Q9400 based on four processors, the second with an AMD PHENOM II X6
In order to find the global minimum of a real function of n variables, a new parallel algorithm of the multi-start and local search type is proposed
Summary
In this paper we consider the following global optimization problem. In this paper we present a parallel algorithm that distributes the computations carried out by Glob across two or more processors. To reduce to a low level the data passing operations between processors, the sequential algorithm is run on each processor, but the parameters of the execution rule are updated either after a fixed number of iterations are completed or straight as soon as new local minimizer is found. The new algorithm has been tested for solving several test functions commonly used in the literature. The numerical experiments have been carried out on two computers equipped with four and six processors; fourteen configurations of the computing resources have been investigated. To evaluate the algorithm performances the speedup and the efficiency are reported for each configuration
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have