CRAFT磁体性能研究平台数据归档系统  

CRAFT magnet performance research platform data archiving system

作　　者：马树良 刘华军 刘方[2] 张舒庆 李童 施毅 郭亮 MA Shuliang;LIU Huajun;LIU Fang;ZHANG Shuqing;LI Tong;SHI Yi;GUO Liang(Institute of Physical Science and Information Technology,Anhui University,Hefei 230601,China;Hefei Institutes of Physical Science,Chinese Academy of Sciences,Hefei 230031,China;University of Science and Technology of China,Hefei 230026,China)

机构地区：[1]安徽大学物质科学与信息技术研究院,合肥230601 [2]中国科学院合肥物质科学研究院,合肥230031 [3]中国科学技术大学,合肥230026

出　　处：《核技术》2023年第2期51-60,共10页Nuclear Techniques

基　　金：中国科学院战略性先导科技专项(No.XDB250020200)资助。

摘　　要：聚变堆主机关键系统的综合研究设施的磁体性能研究平台(Magnet Performance Research Platform,MPRP)是为先进超导磁体实验建立的大型实验平台,其历史数据在海量存储情况下存在检索速度慢的问题。因此,对系统检索速度进行研究并开发了MPRP数据归档系统(MPRP Data Archiving System,MPDAS)。MPDAS设计了EPICS(Experimental Physics and Industrial Control System)数据归档插件并采用MongoDB分片和副本集机制搭建高扩展性数据存储架构。为提高数据检索速度,MPDAS借鉴最近最少使用(Least Recently Used,LRU)、使用频率最低(Least Frequently Used,LFU)、先进先出(First In First Out,FIFO)三种传统缓存替换算法核心思想,基于牛顿冷却定律建立数据温度模型并提出一种综合访问时间、访问频率以及存储顺序的多维度特征数据划分算法。根据数据划分算法标识冷热历史数据实现数据分层存储。MPDAS在查询历史数据时优先访问Redis,根据命中结果和数据完整性选择不同的检索策略。系统测试结果表明:MPDAS功能特征满足设计要求,其搭载的冷热数据划分算法相比FIFO、LRU、LFU在热数据库保存1%历史数据量时的Redis命中率分别提升了38.05%、26.91%和11.06%。通过提高热数据命中率能够直接减少数据检索平均响应时间,MPDAS通过量化历史数据热度并进行冷热划分,有效地提升了系统检索响应速度。[Background]The comprehensive research facility for fusion technology magnet performance research platform(MPRP)is a large-scale experimental platform established for advanced superconducting magnet experiments.The retrieval speed of MPRP historical data is slow due to massive storage.[Purpose]The study aims to develop a MPRP data archiving system(MPDAS)and increase its retrieval speed.[Methods]First of all,the experimental physics and industrial control system(EPICS)data archiving plug-in was designed for MPDAS.Both MongoDB Sharding and Replica Set mechanism were employed to build a highly scalable data storage architecture.Then,the core ideas of three traditional cache replacement algorithms,LRU(least recently used),LFU(least frequently used)and FIFO(first in first out)were drawn by MPDAS to establish a data temperature model based on Newton’s law of cooling.A multi-dimensional feature data partitioning algorithm was implemented to integrate access time,access frequency and storage order,hence the hot and cold historical data were identified to realize data tiered storage.Finally,the retrieval speed of MPDAS was improved by preferentially accessing Redis when querying historical data,and selecting different retrieval strategies based on hit results and data integrity.[Results]The system test results show that the functional characteristics of MPDAS meet the design requirements.Compared with FIFO,LRU,and LFU,the Redis hit rate of the MPDAS when the hot database stores 1%of the historical data is increased by 38.05%,26.91%,and 11.06%respectively.[Conclusions]By increasing the hit rate of hot data,the average response time of data retrieval can be directly reduced.The retrieval response speed of MPDAS is effectively improved by quantifying the heat of historical data and dividing the heat and cold.

关 键 词：历史数据 MONGODB 热数据 PV 分布式集群 

分 类 号：P315.69[天文地球—地震学]