(IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, No. 4, July 2010
Generate the offspring using genetic operatorsconsisting of crossover and mutation. Crossover isa recombination operator that swaps the parts of two parents, two random decisions are made priorto this operation, whether to do it or not and wherethe crossover point is. Mutation gives a goodchance to explore the uncovered search space. Itmutates, or complements some genes in thechromosome of the offspring, so that the newparameter value takes place.5)
Entirely replace the elder generation in the poolwith the newer one and return to step 2. In somecase, the few best elders may be kept away fromreplacement. This is known as elitist strategy. Thecriteria for stopping the revaluation loops are metwhen a (the loop number is over some predefinedpoint or d) the steady last for predetermined times.It is important to realize that GA are stochasticmeaning that there is randomness involved ; mainlyin the initial generation of a random population,random choice of parents, random choice of whichgenes to inherit from which parents, and randomchoice of which genes to mute. Some times howhowever tournament selection is used as a parent orsurvival selection strategy. This ensures that there isbias towards replacing less fit solutions in the parentpopulation by fitter solution from the new generation.2. PARALLEL GENETIC ALGORITHMTwo approaches to parallel genetic algorithms havebeen considered so far.
Standard parallel approach
: In this approach, theevaluation and the reproduction are done in parallel.However, the selection is still done sequentially,because would require a fully connected graph of individuals in the population may be mated.
This approach consists individing the population into equal size sub-populations. Each processor runs the geneticalgorithm on its own sub-population, periodicallyselecting good individuals to send its neighbors andperiodically receiving copies of its neighbors’ goodindividuals to replace bad ones in its own sub-population.
The processor neighborhood, the frequency of exchange and the number of individuals exchangedare adjustable parameters. The standard parallelmodel is not flexible in the sense that thecommunication overhead grows as the square of population size. Therefore, this approach is notadapted to distributed memory architectures, wherethe cost of communications has a great impact on theperformance of parallel programs. In thedecomposition model, the inherent parallelism is notfully exploited as treatment of sub-populations maybe further decomposed. This approach should beconsidered only when the number of availableprocessors is less than the required size of thepopulation. Considering massively parallelarchitectures with numerous processors, we chose afine –grained model, where the population is mappedon a connected processor graph like a grid, oneindividual per processor. We have bisection betweenthe individual set and the processor set. The selectionis done locally in a neighborhood of each individual.Another version to this approach has been alreadyproposed in, where at each generation a hill-climbingalgorithm is executed for each individual in thepopulation. The choice of the neighborhood is theadjustable parameter. To avoid overhead andcomplexity of routing algorithms in paralleldistributed machines, a good choice may be restrictneighborhood to only directly connected individuals.
The parallel genetic algorithm proposed is:
Genetic in parallel a population of randomindividuals.
- Evaluate in parallel eachindividual.
Receive in parallel theindividuals coming from the neighbors.3.
Each individualsreproduces in parallel with the individualspreviously received.
Do in parallel a selection of best local off springs.
We use the speed-up ratio as a metric for theperformance of the parallel genetic algorithm. Thespeed up ratio S is defined as
is theexecution time on a single processor and
corresponds to execution time for a
processorsimplementation.3. RELATED WORKThere are mainly two types of scheduling namelythe system level scheduling and the application levelscheduling. The scheduling system will analyze theload situation of every node and select one node torun the job. The scheduling policy is to optimize thetotal performance of the whole system. If the systemis heavily loaded, the scheduling system has torealize the load balancing and increase the throughput