A Multi-Population Two-Stage Random Key Genetic Algorithm For Cloud Workflow Scheduling Optimization

Cloud computing is a new type of commercial.It virtualizes various hardware facilities and provides users with efficient and scalable computing services in a form of billing based on usage.With the continuous improvement of productivity,scientific research and production activities require very powerful computing capabilities to analyze and calculate the amount of data.It is difficult for the IT architecture of various industries,since a single organization is difficult to achieve the required computing requirements.The advantage of cloud computing is that there is no need to purchase IT infrastructure,build a private computing platform,and manage and upgrade software and hardware resources and invest high costs and labor costs.Cloud workflow,as its name implies,is the application of cloud work-related technologies to cloud computing,which can make workflow more efficient,cloud computing more efficient and more reasonable allocation of resources.This article first introduces related basic concepts,establishes a cloud workflow scheduling optimization model,and explains different types of algorithms.Since the intelligent computing method has the problem of incomplete search space or low search efficiency when solving the cloud workflow scheduling problem,a multi-group two-stage random key genetic algorithm(MTRKGA)is proposed,which discusses its encoding,two-stage decoding,population initialization,crossover operation,individual improvement,population migration and evolution strategies,and the overall algorithm framework.The algorithm uses n-dimensional real number coding when coding,so any coding scheme can find a corresponding scheduling optimization scheme,so the search space is complete.Two chromsomes based on the heterogeneous earliest finish time(HEFT)and the dynamic HEFT(DHEFT)are seeded into the initialization population,which improves the quality of the initial population and shortens the search time.When generating new populations,not only traditional crossover methods are used,but individual migration methods are also used to replace mutations,which can destroy individuals to a greater extent and ensure population diversity.In the evolution process,a multiple population co-evolution strategy is used,which make the sub-populations escape the local optimum and accelerate the convergence of the sub-populations.In addition,a two-stage co-evolution strategy is also used.The first stage uses heuristic-based earliest finish time decoding to make the algorithm converge to to the vicinity of the optimal solution as soon as possible.In the second stage,the decoding method based on insertion mode,the forward and backward decoding improvement(FBI)and load balancing improvement(LDI)are used to expand the search,which enhances the algorithm optimization ability and search efficiency.Finally,in order to verify the effectiveness of the proposed MTRKGA,the algorithm is implemented using C++.The proposed MTRKGA is compared with some current existing clasic algorithms(HEFT^[17],HGA^[59],LWSGA^[47],CGA^[45],etc.)in various workflow cases.The experimental results show MTRKGA not only has high search efficiency and can find better solutions in the same time,but also has a certain degree of robustness.