Scientific workflow tasks are an efficient model of scientific computing task.With the increasing complexity of scientific computing,the demand for computing resources is also increasing dramatically.As a flexible computing service mode,cloud computing meets the computing resource demands of workflow tasks,and more and more workflow tasks are being transferred to the cloud computing for execution.Therefore,workflow task scheduling algorithms in cloud computing environment have become one of the hot research directions.As a typical representative of heuristic algorithms,particle swarm optimization algorithm has a simple structure and is easy to understand,but the algorithm itself has problems such as premature convergence and low population diversity during the running process,which can easily lead to the local optimal solution.In addition,current workflow tasks still have problems such as single optimization goal and high waste ratio in cloud spending during the scheduling process.In order to solve these problems,the main contents of this paper are as follows:Firstly,in view of the single optimization objective of workflow task scheduling,which leads to the loss of the optimization performance of the relevant parameters of the system(such as load balancing and cost,etc.),this paper designs a multi-objective optimization model based on the traditional single objective optimization model,which considers the completion time,load balancing and execution cost,and flexibly adjusts the proportion of each objective in the overall optimization in a weighted way.To solve this model,an improved multi-swarm particle swarm scheduling algorithm IMSPSO is proposed.Firstly,the particles are sorted and grouped according to their fitness value,and the swarm renewal strategy is designed to improve the population diversity of the algorithm throughout the iteration process.Secondly,the division of labor and cooperation between the swarms are set,with particles in the master and slave swarms using different updating methods,and the local search accuracy and global search ability of the algorithm are taken into account.Finally,the improved algorithm is compared with other improved PSO algorithms.The results show that the proposed algorithm achieves better results in multi-objective optimization.Secondly,to address the issue of the high waste ratio in the current cloud computing spending,this paper reduces waste from the angle of improving the utilization of virtual machine resources,and designs a model with completion time and resource utilization as the optimization target.In order to solve the model,a scheduling algorithm CS-PSO that combines cuckoo search algorithm and particle swarm optimization is proposed.Firstly,a criterion for judging whether the particle swarm algorithm is trapped in a local optimal solution is designed,and the particles are checked for local optimization in each iteration.Secondly,optimization of Cuckoo search algorithm: First,aiming at the problem that cuckoo search algorithm parameters are fixed,it is changed dynamically with the number of algorithm iterations;Secondly,the nest position update method of Cuckoo search algorithm draws on the idea of particle swarm optimization algorithm to optimize the nest position update according to the global optimal nest.Finally,the particles trapped in the local optimization are identified,and the positions are updated using the improved cuckoo search algorithm to improve the global search capability of the algorithm.Experimental results comparing the improved algorithm with some heuristic algorithms show that the proposed algorithm effectively reduces waste in cloud computing.Thirdly,based on the above research,this paper designs a workflow task scheduling system in a cloud environment.Combining the two workflow scheduling algorithms proposed above with the actual application requirements,the main functions and module division of the system are designed.The system adopts the development method of front-end and back-end separation,which is convenient for later business expansion.In addition,through the demonstration of actual cases,it can be seen that the system data display methods are diverse,the interface is simple,the operation is simple,and it has a good user experience. |