当前位置:首页 > 报告详情

计算存储案例研究:真实用户部署.pdf

上传人: a****d 编号:185012 2024-10-07 14页 655.72KB

1、1|2024 SNIA.All Rights Reserved.Computational Storage Case Studies:Real User DeploymentsJB Baker,ScaleFlux3|2024 SNIA.All Rights Reserved.Agenda Promises of Computational Storage Deployment Examples:HTAP Database Cloud Parallel File System Relational Database Deployment Challenges4|2024 SNIA.All Rig

2、hts Reserved.Promises of Computational Storage5|2024 SNIA.All Rights Reserved.Promises of Computational StorageMoving compute functions to the data instead of the data to a CPU to improve:SustainabilityEfficiencyPerformanceTCOSecurity6|2024 SNIA.All Rights Reserved.Deployment Examples7|2024 SNIA.All

3、 Rights Reserved.Computational Storage for HTAP Database Deployment Characteristics Hybrid Transactional Analytical Processing Database Massive scale cloud service 1000s of customers across 200+countries and regions Desired Outcomes Reduce Cost of infrastructure Reduce Power consumption Reduce Maint

4、enance&Complexity All while meeting existing SLAs8|2024 SNIA.All Rights Reserved.Computational Storage for HTAP DatabaseUsing NVMe CSDs for Da ta Com pressionBa seline With Ordina ry NVMe SSDs65 Ca binets650 Servers7800 SSDs60%lower:Power Spa ce Cost26 Ca binets260 Servers3125 SSDsEquipm ent NeededI

5、m provem ent9|2024 SNIA.All Rights Reserved.Computational Storage for Cloud Parallel File System Deployment Characteristics Massively parallel file system Microsecond access latencies required Serving multiple workloads:AI Training,EDA simulation,CG rendering,and more Triple Replication Desired Outc

6、omes Reduce Cost of infrastructure Reduce TCO Keep or improve Latency SLAs Scale performance with number of users10|2024 SNIA.All Rights Reserved.Computational Storage for Cloud Parallel File SystemCopy ACopy BCopy CGPU NodeUsing NVMe CSDs for Da ta Com pressionBa seline With Ord

word格式文档无特别注明外均可编辑修改,预览文件经过压缩,下载原文更清晰!
三个皮匠报告文库所有资源均是客户上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作商用。
本文主要介绍了计算存储(Computational Storage)的实际部署案例,强调了将计算功能移至数据存储中,以提升能效、性能和可持续性,同时减少基础设施成本、功耗和维护复杂性。关键数据包括:使用NVMe计算存储设备(CSDs)比标准NVMe SSD节能60%,节省空间60%,降低成本50%;在云并行文件系统中,采用CSDs后,存储成本降低,总体拥有成本(TCO)减少,访问延迟保持或优化,且能随用户数扩展性能;在关系型数据库服务中,采用CSDs的集群成本降低50%,节点数减少75%。尽管存在采用门槛,如对NVMe和存储技术的专业知识需求、容量扩展管理、应用与硬件团队的目标竞争以及计算存储功能的局限性等,但通过开发服务器管理插件、软件定义存储(SDS)集成和市场教育,这些障碍正逐步被克服。
计算存储如何提高基础设施成本效益? 计算存储在云并行文件系统中的应用如何实现低延迟和高性能? 计算存储在关系型数据库中的部署挑战有哪些?如何克服?
客服
商务合作
小程序
服务号
折叠