数据密集型超算现状、挑战以及未来发展趋势  

Status,Challenges,and Trends of Data-Intensive Supercomputing

作　　者：魏嘉 陈默[2] 王龙翔[1] 任沛 雷雨佳 屈俞岐 蒋骐羽 董小社[1] 伍卫国[1] 张凯丽 张兴军[1] WEI Jia;CHEN Mo;WANG Longxiang;REN Pei;LEI Yujia;QU Yuqi;JIANG Qiyu;DONG Xiaoshe;WU Weiguo;ZHANG Kaili;ZHANG Xingjun(School of Computer Science and Technology,Xi’an Jiaotong University,Xi’an,Shaanxi 710049,China;Huawei technologies Co.Ltd.,ShenZhen,Guangdong 518129,China)

机构地区：[1]西安交通大学,计算机学院,陕西西安710049 [2]华为技术有限公司,广东深圳518129

出　　处：《数据与计算发展前沿》2023年第3期66-91,共26页Frontiers of Data & Computing

基　　金：国家重点研发计划(2016YFB0200902)。

摘　　要：[目的]对数据密集型超算的发展历程、主流系统结构、典型应用和计算、存储子系统进行较全面的梳理,指出未来发展趋势,为后续数据密集型超算系统优化提供参考。[方法]本文首先梳理了数据密集型超算中的关键概念,分析阐述现有平台对数据密集型应用的支持程度。说明了科学界和工业界对数据密集型应用的实际需求情况。并对数据密集型超算的未来发展趋势和面临的潜在挑战进行展望,建立了超算系统评测模型。[结果]相关研究人员和从业者可从本文快速了解到超算技术的关键概念及发展状况,精准捕捉当下与未来数据密集型超算研究热点和亟待解决的关键问题。[结论]数据密集型超算存储系统面临的复杂数据类型优化、混合负载优化、多协议支持与互通等将会成为未来一段时间内研究和发展的热点问题。[Objective]This paper is to provide a comprehensive and systematic overview of the development history,mainstream system architecture,typical applications,and computation and storage subsystems of data-intensive supercomputing,point out the future development trend,and provide references for further data-intensive supercomputing optimization.[Methods]This paper first sorts out the key concepts of data-intensive supercomputing and analyzes the support to the data-intensive applications by existing platforms.Then the real demand for data-intensive applications from the mainstream academic and industrial communities are illustrated.Finally,the future trends and potential challenges of data-intensive supercomputing are discussed and a corresponding supercomputing system evaluation model is developed.[Results]Relevant researchers and practitioners can quickly understand the key concepts and development status of supercomputing technology from this paper,and precisely capture the current and future data-intensive supercomputing research hotspots and key problems that need to be solved.[Conclusions]The problems such as the optimization on complex data type and mixed workload,and multi-protocol support and interoperability which are faced by the data-intensive supercomputing storage systems will become hot research and development issues in the coming years.

关 键 词：数据密集型超算 I/O密集型超算 高性能数据分析 并行处理系统 超算存储系统 

分 类 号：TP338.4[自动化与计算机技术—计算机系统结构]