Research On Cloud Computing Resource Scheduling Based On Improved Shuffled Frog Leaping Algroithm

Cloud computing has become a well-known hot word of science and technology, but also become a highlight research question in the field of academia and industry. Cloud computing virtualizes all resources, puts them in a resource pool, and schedules these resources for the use of each task transparently, so the virtualization mapping between application layer and virtual resource layer is the key part of cloud computing. Resource allocation is to find an optimization scheme to implement a reasonable mapping between tasks and resources. How to find a reasonable assignment scheme is critical for the resource scheduling. In recent years, many researchers have also studied the problem and proposed several techniques. Since SFLA(Shuffled Frog Leaping Algorithm) appeared a short time, many of the theoretical basis and parameter settings is not very mature, so the researches of the cloud computing resource scheduling about SFLA is not too much. some of the existing algorithms and improved algorithms, to some extent, there are some shortcomings and deficiencies. There are not some researches about improving the fitness and quality of the initial frog population. The worst individual moving step is relatively simple. The population is very easy to fall into local optimum, like these problems, and so on.Aiming at the problems existed in the cloud computing resource scheduling about SFLA application, in order to improve the convergence speed, convergence accuracy and optimization capabilities, the article has put forward the improved SFLA. The main works of improvements have three modules related. In population initialization, taking the classical Min-Min algorithm and random method to initial the frog population, greatly improve the quality of the initial population, the task completion time can be indirectly shortened; in local search strategy, the original algorithm only took single moving step formula, this may be result in the worst frog still is the worst individual of subgroups after updating one time. so introducing the concept of average fitness value, in the subgroups, by comparing the fitness value of worst frog with the average fitness value to set the moving step formula to determine the moving space of worst frog, than updating the position of worst frog, this can avoid the blind local search, make the converge speed more faster and optimization capabilities better; in the global mixing operation, take the thought of crossover on genetic algorithm into consideration. When the entire frog individuals mix into a big group, they will exchange of culture and information. By setting the appropriate crossover control parameters, it will operate the best frog individual of local and global. Reserving the better frog individual after mixing, to a certain extent, this method will avoid falling into local optimum and the progeny will own parent’s good genes, can also improve the quality of solution. In real life, there are in great demand tasks of user and the numbers of resources are limited, so the improved algorithm of short time and quick is more appropriate to the circumstances of this limited condition.In this paper, simulated experiments in Cloud Sim that is cloud computing simulation platform. As SFLA is relatively new concept, the parameter settings do not like mature genetic algorithm. A large number of experiments have to be repeated to do, in limited circumstances, getting the better the group of population size. Then, under such a grouping criteria, changed the value of other variables to validate algorithm performance. The experimental results show that, at a certain number of resources, the tasks are increasing and at a certain number of tasks, changing the number of resources, compared to the genetic algorithm and particle swarm optimization, the improved leapfrog algorithm has better convergence speed and optimization capabilities. In generally, resource scheduling has more efficiency. It is more suitable for the situation of more tasks and less resources, proves the feasibility of the algorithm.