Study On Resources Scheduling Based On Particle Swarm Optimization Algorithm In Cloud Computing

In order to meet the enormous needs of users especially when dealing with massive data and tasks,cloud resource scheduling optimization algorithm is directly related to the allocation of computing nodes and tasks,thereby affects the overall performance of the cloud platform,and is a key link to enhance cloud services.For cloud resource scheduling problem,it is a very complex NP problem,when the system filled with certain computing scale,it is difficult to solve these problems using linear programming,simplex method,Newton method and other conventional methods.But for genetic algorithm,colony algorithm,differential evolution algorithm and other intelligent algorithms,it is more easier to get better results.However,these algorithms tend to fall into local convergence at the later stage of the iteration,which makes the improvement of the optimization results less than expected.Particle swarm optimization(PSO)is widely used in the solution of such complex NP problems due to its simple structure and strong optimization ability.Although a lot of researches on task scheduling and resource allocation are existed in cloud computing,but as for optimization index,such as shortening task completing time,improving the utilization rate of the load and reducing the task cost,it is usually to choose one or two of three indicators as the research focus,and not consider three indicators totally.Based on the problems above,this thesis begins researches in three aspects:1.As for the problems existed in LDW-PSO algorithm,First of all,based on the linear decreasing inertia weight,making the inertia weight significantly increased with constant disturbance,in order to jump out of local search and global search,so as to prevent local convergence;At the same time,in order to avoid that particles in particle swarm algorithm heightly gather around the optimal particle as much as possible at the later stage of the algorithm,change the inertia weight adaptively mixed with random particles to some extent,so that it can better maintain the diversity of population.Then,based on the above improvement strategy,add particle swarm segmentation strategy,the whole iteration cycle is divided into two parts,and suitable inertia weight interval is defined,each part is updated in different ways,ensuring that all values are selected in suitable intervals,and the convergence of particle swarm is enhanced.2.The definition of evaluation index in cloud resource scheduling: total task completion time,total load balancing,total stask completion costs,set a reasonable objective function,at the same time,the screening mechanism is added in,in thecondition of the same load balancing value on some node,select the node which has the characteristic of shorter execution time and low cost to execute tasks,considering in three optimizing index.3.Under the Matlab2010 a GUI platform,typical example functions are used to verify accuracy of the DLPSO algorithm,combined with the cloud resource scheduling model,also compared with other particle swarm algorithm,The outcome of the experiments demonstrate that the DLPSO algorithm has higher accuracy,the single objective optimization strategy of task execution time is shorter.In order to verify the effect of DLPSO algorithm with three optimization index in the cloud environment,based on Cloudsim cloud simulation platform,a variety of other particles swarm optimization algorithms are used for comparison,simulation results show that DLPSO algorithm is more easy to get the accurate optimal solution,shorten the task completion time,improve the load balance,reduce total cost.