Databases Reference
In-Depth Information
Fig. 8.9.
Master-slave model of parallelization.
by re-considering the representation of the individuals, redesigning the
reproduction operators, balancing between the population size and the
number of generations, introducing suitable local search operators, etc.
Parallel GAs are typically classified into three classes according to their
approach on parallelization: fitness level, population level and individual
level parallelization. Hierarchical or hybrid parallelization models are often
considered as the fourth class.
Fitness level:
In the fitness level parallelization a master process
distributes the evaluation of the individuals to slave processes. The master
process performs all the other tasksoftheGAshowninFig.8.9.This
kind of parallelization is useful when the fitness value calculation is time-
consuming so that the data transfer between the master and the slaves will
not become a bottleneck of the system.
Population level:
In the population level parallelization a complete
GA including an entire population is placed at each process and an
interconnection network between the processes is constructed, which is well
described in Fig. 8.10.
Individual level:
Parallelization can also be realized at individual level.
Commonly, the individuals are placed in a grid. Each grid node contains a
single individual.
