Research On Many-Core Oriented Workflow Scheduling Under Cloud Computing

As an emerging computing service model,cloud computing's scalability and on-demand resource allocation make it a huge driving force for development in academia and industry.Many-core computing resources can be used,especially in the field of scientific computing.Many complex application tasks are deployed in the form of workflows.In the era of big data,the speed of data accumulation is far greater than its processing speed.Traditional workflow scheduling has been unable to cope with the ever-growing data size and analysis complexity.How to deal with the relationship between workflow and many-core computing resources under cloud computing is particularly important,and related research work has only just begun.Therefore,many-core oriented workflow scheduling under cloud computing has theoretical research significance and practical value.The research in this thesis is based on cloud computing,many-core computing and workflow analysis.It explores the workflow scheduling algorithm,tries to solve or partially solve the common problems in the current classical algorithms,and applies the proposed algorithm to workflow scheduling framework.The main contributions of this thesis 's research work are as follows:First,consider the resource scheduling and user computing needs of many-core platforms in cloud computing environment.Combining the ideas of genetic algorithms,this thesis proposes an adaptive re-evolutionary cloud workflow scheduling algorithm(abbreviated as AGAR algorithm)under constraint conditions,using the modified mean value calculation.The relevant parameters and the adaptive penalty function modify the fitness.We introduce a re-evolutionary strategy,and give an evaluation index that integrates time constraints and cost constraints to find the optimal scheduling strategy.Finally,it minimizes the implementation cost of the scheduling goal under the premise of meets the user's calculated requirements.Based on previous work,this experiment selected four scientific work as test data.It mainly focused on user-submitted dependent calculation tasks,analyzed and compared the performance of multiple scheduling algorithms under comprehensive evaluation indicators,and demonstrated that the proposed scheduling strategy of AGAR algorithm has the excellent performance.Second,study the cloud workflow scheduling framework.Analyzes the computing needs of cloud workflows,data processing needs,and the challenges faced by cloud workflow scheduling.It analyzes the problems of multiple types of many-core computing resources which are difficult to use and collaborate.The thesis proposes a four-tier cloud workflow scheduling framework and design support primitive collection.Combining dynamic scripting language and cloud service features,cloud workflow scheduling framework is integrated in the dynamic scripting support subsystem of transparent cloud environment.