The Research On The Kriging Based Cloud Task Scheduling Method And The Engineering Optimization Cloud Platform Design

Cloud computing can provide the high quality and high availability services to users in "pay-as-you-go" ways with the high performance hardware and software resources in datacenters, the "offering computing resources as services" novel computing strategy has been applied in many fields. Building the cloud platform for engineering optimization lacks in mature techniques as demanded by the complex application backgrounds in engineering optimization problems. Firstly, high performance engineering optimization method must be accomplished for cloud application development as for the cost saving of the users in the cloud platform; Secondly, the traditional cloud task scheduling methods are too universal and can barely adapt the special needs for engineering optimization cloud tasks; Thirdly, the intensive characteristics of the response time and the computing cost for the engineering optimization cloud applications have called for new requirements for the usage of the users as well as the allocation of the platform resources.To solve the above problems, the following work has been done in this dissertation:1. A parallel optimization method based on Entropy-based Expected Improvement criterion has been proposed. By introducing entropy theory into weighted expected improvement criterion, an optimal weight could be gotat every iteration which will has both the maximum expected improvement and the optimal weight characters simultaneously. With parallel computing technology, the optimization procedure can be divided into large particles by separating the samples and weighted parameters through parallel processes. The high performance parallel optimization method using EEI criterion can not only guarantee a more precise optimal solution but also high parallel speedup which shortening the optimization procedure.2. A Kriging surrogate model based dynamic cloud task scheduling method is proposed. Based on the steady period and transition period characters existing in the microcosmic of the engineering optimization cloud applications, a Kriging based optimization solution is developed for steady periods. The resource combinations would be the design variables and the response time and the computing cost would be the objective function, the optimal resource scheduling strategy will be obtained after optimization. Besides assuring the cloud tasks for being computed in shorten time durations, the method will also decrease the resource consumption of the users and increase the utilization of the resources approaching the maximization.3. A Kriging based cloud task forecasting and allocation method is proposed and an engineering optimization cloud platform is designed. Based on the macro computing characters in engineering cloud applications, an optimization model can be developed by letting the application parts as well as the computing resource parts be the design variables and the cloud task response time and the computing cost be the objective functions. With the help of the Kriging model between the design variables and the response time, the response time of the new cloud task can be forecasted, and the computing resource allocation will be a constrained minimum optimization problem with certain computing cost is provided. The forecasting of the response time for cloud tasks can avoid the unnecessary waiting process by the users, the resource allocation strategy can avoid the blindness of using the platform resources according to users experience and help them arrange their computing tasks properly. Cloud modules are constructed by virtualization technologies, key features are designed, and the cloud platform for engineering optimization applications have been finally built up.