高能物理环境中混合存储系统的设计与优化

发布时间：2018-06-01 21:52

本文选题：海量存储系统 + 高能物理　；参考：《计算机科学》2017年10期

【摘要】：高能物理是典型的数据密集型计算环境,数据处理包括模拟计算、重建计算以及物理分析。其中大文件计算占据较大比重,并且高能物理文件访问模式以跳读为主,因此大文件的高速访问成为整个系统性能的重要影响因素。首先剖析传统高能物理计算环境的典型架构及其文件访问模式的特点,介绍混合存储模式在高能物理计算环境中的优势,总结其数据访问方式的特点,对其各种读写方式进行数据测试;然后提出针对该环境的混合存储系统的部署设计和优化,使该环境下的数据读写性能得到明显提高;同时将成本因素考虑到系统设计中,实现了一个低成本高性能的存储系统。测试表明,混合存储系统在高能物理等大数据存储系统中具有高效的I/O性能。文中全面分析了影响其性能的各种因素,实现了最优化配置的低成本高性能混合存储系统,并对该系统的未来发展趋势进行了分析和展望。
[Abstract]:High energy physics is a typical data intensive computing environment. Data processing includes analog computing, reconstruction calculation and physical analysis. Large file computing occupies a large proportion, and high energy physical file access mode is dominated by skipping, so the high speed access of large files has become an important factor in the whole system performance. The typical architecture of high energy physics computing environment and the characteristics of file access mode are introduced. The advantages of hybrid storage mode in high energy physical computing environment are introduced, the characteristics of the data access mode are summarized, and the data of various reading and writing methods are tested. Then, the design and optimization of the mixed storage system for this environment are put forward, so that the design and optimization of the hybrid storage system for this environment are proposed. The performance of data reading and writing in the environment has been greatly improved. At the same time, the cost factor is considered in the system design, and a low cost and high performance storage system has been realized. The test shows that the hybrid storage system has high performance I/O performance in high energy physics and other large data storage systems. The low-cost and high-performance hybrid storage system with optimized configuration is analyzed, and the future development trend of the system is analyzed and prospected.
【作者单位】：中国科学院大学物理科学学院;中国科学院高能物理研究所;
【基金】：国家自然科学基金项目:高能物理实验的大规模离线数据存储技术研究(11575223);国家自然科学基金项目:基于SDN的高能物理云数据中心弹性网络关键技术研究与应用(11605224)资助国家重点研发计划项目:科学大数据管理系统(2016YFB1000605)
【分类号】：TP333

【相似文献】