| Recent advances in theoretical,computational and experimental materials science and engineering present not only the promise to speeding up the pace at which novel materials are discovered,but also to decrease the time required to bring these discoveries as new products to market.The utilization of high-throughput density functional theory(DFT)calculations for screening of new materials and conducting fundamental research offers also interesting design opportunities for materials science and materials innovation.High-throughput DFT usually involves computations on tens or even hundreds of thousands of compounds,and such a change of scale demands new computing capabilities and data management methodologies.So far,to ensure timely computation,such high-throughput DFT material simulations have had to run on dedicated high-performance computers(HPCs).However,as the number of jobs and the complexity of the simulations needed to generate and analyze materials data increase,it becomes a challenge for HPC environments to solve the given scientific problem within a reasonable amount of time.The emergence of the latest computing paradigms,Grid and Cloudcomputing,has brought about a sweeping change in both the scientific and IT communities.On the one hand,Grid computing allows heterogeneous resources ranging from personal computers to supercomputers to be accessed and shared in a secure and flexible manner in order to solve large-scale problems arising in scientific and engineering domains.On the other hand,Cloud computing is a further development of the Grid,brings with it tremendous opportunities to host and run real-world applications from different domains at relatively low costs without the need of owning any IT infrastructure.Besides,it can also deliver various hardware resources as services to consumers over the internet.Considering the obstacles of running high-throughput DFT calculations over HPCs,this thesis aims to investigate running high-throughput DFT material simulations over both Grid and Cloud computing environments in order to take full advantage of the large-scale heterogeneous collection of geographically distributed and interconnected autonomous resources provided by the Grid,as well as benefiting from the possibility of acquiring huge amounts of Cloud resources in a dynamic and elastic way.In this thesis,new algorithms have been designed to handle the job scheduling issues in Grid and Cloud computing environments.In more detail,the thesis makes the following key contributions:1)An effective job scheduling algorithm,called two choices scheduling algorithm(TCSA),has been proposed to dynamically allocate jobs to resources so as to minimize the jobs execution time and maximize resource utilization in Grid environment.2)An improved particle swarm optimization(PSO)algorithm has been developed to solve the job scheduling problem in Grid environment.The developed PSO algorithm aims at minimizing the makespan and flowtime of job scheduling simultaneously.3)A randomization-based load balancing algorithm has been designed to solve the job scheduling problem in Cloud computing environment.The algorithm seeks to minimize jobs execution time and maximize resource utilization by distributing workloads among resources evenly in the Cloud.4)A simplified version of particle swarm optimization(PSO)algorithm has been proposed to solve the job scheduling problem in Cloud computing environment,with makespan as an objective.To evaluate the performance of the proposed algorithms,we have compared the presented job scheduling algorithms with several existing state-of-the-art scheduling algorithms through conducting several simulation experiments under different scenarios.The experimental results show that our designed algorithms work very well and they are able to find optimal or near-optimal solutions within a reasonable amount of time.In addition,our approaches significantly outperform the other compared algorithms in terms of the mentioned objectives,especially when the scheduling problem becomes too complex or too large. |
