Research On The Improved Flower Pollination Algorithm For Multi-objective Workflow Scheduling In Cloud Computing

Makespan,scheduling cost,resource utilization,and energy consumption are the key evaluation criteria for workflow scheduling.However,compared to traditional distributed systems such as grid computing,the factors that affect these evaluation criteria increased due to cloud computing's complex and changeable characteristics.If this evaluation criteria cannot be balanced,not only will the user's experience be reduced and the cost is increased,but the service provider's market competitiveness also be reduced.Therefore,how to optimize the makespan,scheduling cost,resource utilization,and energy consumption are important for workflow scheduling in the cloud environment.Most of the study is contributed to reduce the cost under a deadline or reduce the makespan under budget,but there is still room for research on multi-objective workflow scheduling optimization issues.In view of the shortcomings of the past study,the main contents of this paper are as follows:Aiming at the problem that the optimization goal of existed workflow scheduling methods is singularization,the dissertation firstly proposes a multi-objective optimization model for workflow scheduling in cloud computing,the scheduling goals are to reduce makespan,scheduling costs and energy consumption,meanwhile,improve resource utilization.Then the dissertation improves the flower pollination algorithm from a numerical optimization algorithm to a combinatorial optimization algorithm,which is suitable for workflow scheduling.Because meta-heuristic algorithms are susceptible to local optimization,differential flower pollination multi-objective scheduling algorithm(DMFPA)is proposed.By improving the transition probability of the flower pollination algorithm and combining the differential algorithm to improve the diversity of the population,it's global search capability and convergence speed has significantly improved and can find approximately optimal solutions in polynomial time.Aiming at the problem that the task execution time is uncertain due to the complexity of the cloud environment,the dissertation models the uncertainty of the task execution time through the Z-Number.It's improved the population's initialization,transfer probability,and pollination operations on the basic flower pollination algorithm,and then a hybrid pollination multi-objective workflow scheduling algorithm(HFPA)is proposed.The specific improvement is based on the optimal strategy of virtual machine allocation rules for population initialization,adaptive switch probability,and two-way learning local pollination and greedy global pollination,this can avoid the generatio n of infeasible solutions,and as the number of iterations increases,maintain the balance between switch probability and population diversity.In order to facilitate the comparison of the running results of different scheduling algorithms,a cloud computing workflow scheduling simulation platform was developed based on the open-source framework Workflow Sim.The platform has the characteristics of ease of use and scalability.It integrates two types of model scheduling algorithms,such as the determined clo ud environment and uncertain cloud environment.By selecting different workflow execution files,the operation results of the algorithm can be obtained more conveniently.The comparison between the performance of different algorithms provides an experimental basis.