CAREER: Flashing Up Data Centers: An Orchestrated Design for Flash-based Distributed Storage Systems

职业：闪存数据中心：基于闪存的分布式存储系统的精心设计

• 批准号: 1453705
• 负责人: Feng Chen
• 金额: $ 54万
• 依托单位: Louisiana State University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2015
• 资助国家: 美国
• 起止时间: 2015-02-01 至 2021-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1453705&HistoricalAwards=false
• 关键词: CAREER; Flashing; Up; Data; Centers

A large-scale deployment of flash devices into data centers can greatly improve the overall system performance and reduce the rapidly growing management cost (e.g., power, cooling, staffing, floor space). Despite the technical merits promised, such a grand technical transition fundamentally changes the long-held system design assumption for a disk-based storage and will inevitably bring major critical challenges in a real-world practice. For example, underutilization of flash space would cause huge economic loss; premature device wear-out may result in catastrophic data corruption; unbalanced system could bring severe resource contention; unoptimized applications may not receive anticipated benefits; and many others. This project aims to address these challenges. Research will be conducted to develop a cohesive design approach to providing an orchestrated whole-system optimization. By revisiting the entire storage hierarchy, from hardware, operating system, cluster middleware, to applications, the team will redesign the device architecture to enable an organic integration of flash devices as integral elements in a huge flash storage system, create a flash-based distributed storage service with optimized resource utilization and guaranteed data reliability. Furthermore, a set of key data center applications will be enhanced to fully exploit the great potential of the flash technology. As part of this CAREER project, the team will also seek influence to the industry, contribution to curriculum, and outreach to local area under represented students.

将闪存设备大规模部署到数据中心可以极大地提高整体系统性能，并降低快速增长的管理成本(例如电力、冷却、人员配备、占地面积)。尽管具有技术优势，但如此重大的技术过渡从根本上改变了长期以来对基于磁盘的存储的系统设计假设，并将不可避免地在实际实践中带来重大的关键挑战。例如，闪存空间利用不足会造成巨大的经济损失；设备过早损坏可能会导致灾难性的数据损坏；系统不平衡可能会带来严重的资源争用；未优化的应用程序可能得不到预期的好处等等。该项目旨在应对这些挑战。将进行研究，以开发一种协调一致的设计方法，以提供协调一致的全系统优化。通过重新考察从硬件、操作系统、集群中间件到应用程序的整个存储层次结构，该团队将重新设计设备体系结构，以实现闪存设备作为庞大闪存存储系统中不可或缺的元素的有机集成，创建基于闪存的分布式存储服务，并优化资源利用率和保证数据可靠性。此外，还将增强一系列关键数据中心应用，以充分挖掘闪存技术的巨大潜力。作为这一职业项目的一部分，该团队还将寻求对行业的影响力，对课程的贡献，以及对有代表性的学生所在地区的外联。

项目成果

SlimCache: An Efficient Data Compression Scheme for Flash-based Key-value Caching

SlimCache：一种基于闪存的键值缓存的高效数据压缩方案

• DOI: 10.1145/3383124
• 发表时间: 2020
• 期刊: ACM Transactions on Storage
• 影响因子: 1.7
• 作者: Jia, Yichen;Shao, Zili;Chen, Feng
• 通讯作者: Chen, Feng

Kill Two Birds with One Stone: Auto-tuning RocksDB for High Bandwidth and Low Latency

• DOI: 10.1109/icdcs47774.2020.00113
• 发表时间: 2020-11
• 期刊: 2020 IEEE 40th International Conference on Distributed Computing Systems (ICDCS)
• 作者: Yichen Jia;Feng Chen

From Flash to 3D XPoint: Performance Bottlenecks and Potentials in RocksDB with Storage Evolution

从闪存到 3D XPoint：RocksDB 存储演进的性能瓶颈和潜力

• DOI: 10.1109/ispass48437.2020.00034
• 发表时间: 2020
• 期刊: Proceedings of 2020 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS
• 作者: Jia, Yichen;Chen, Feng
• 通讯作者: Chen, Feng