摘要
Abstract
With the rapid development of machine learning and big data technology,the demand for computing resources in the field of meteorology is increasing exponentially.Under the condition of limited hardware resources,how to improve the computing power utiliza-tion efficiency of high-performance computing cluster(HPC)has become a key research topic.Based on the architecture and business characteristics of a provincial meteorological HPC cluster,this paper puts forward a multi-dimensional collaborative optimization frame-work,and conducts research from four core levels:hardware resource scheduling,system configuration and tuning,dynamic task manage-ment and compilation efficiency improvement.By building a hierarchical optimization model,the performance of storage I/O throughput,job scheduling efficiency,and compilation and execution efficiency are significantly improved.The research results provide a reusable technical path for the efficiency optimization of HPC system in the meteorological field.关键词
高性能计算/算力优化/气象应用/异构资源调度Key words
High performance computing/Computing power optimization/Meteorological application/Heterogeneous resource scheduling分类
信息技术与安全科学
