Research On Workflow Scheduling And Offloading Algorithm In Cloud And Edge Computing

With the rapid development of the Internet of Things,more and more applications appear in the form of workflow.How to use cloud computing and edge computing to schedule and offload workflows is the key to improve the performance of mobile devices and the response time of applications.Firstly,workflow scheduling problem in cloud environment is the initial research of this thesis.Then this thesis further studies the workflow offloading problem in edge computing.Current research on workflow scheduling in cloud computing,very few of them have considered the fairness among workflow tasks when considering the overall goals such as time,cost and energy.To address such a problem,this thesis proposes a workflow scheduling algorithm based on stable matching game theory.In addition,in order to improve the efficiency of algorithm,two local optimization strategies based on critical path and task duplication are added to minimize workflow makespan and ensure the fairness among tasks.Finally,the effectiveness of the proposed algorithm and strategies are verified through several groups of comparative experiments.Due to the shortage of cloud computing in big data transmission,real-time data processing,cost and energy consumption,edge computing is needed to deal with timesensitive and computation-intensive workflows generated by mobile devices.In this thesis,a face recognition application is taken as an example to study the problem of multi-objective and multi-workflow offloading in the edge environment,and a workflow offloading algorithm based on SPEA2 is proposed to minimize the execution cost and execution energy consumption of multi-workflow.During initialization,the deadline of workflow is assigned to each task as a sub-deadline to obtain more initial solutions that satisfy the constraint of deadline.In the crossover operator,the adaptive clustering algorithm is used to help each individual find a suitable mate.According to the historical information,adaptive crossover and mutation probability are introduced to guide the evolution direction of the population.Finally,the effectiveness of the proposed algorithm and strategies are verified through several groups of comparative experiments.