CAREER: Distributed Protocols and Primitives Optimized for RDMA Read/Write Operations in Data Management Systems

职业：针对数据管理系统中的 RDMA 读/写操作进行优化的分布式协议和原语

• 批准号: 2045976
• 负责人: Roberto Palmieri
• 金额: $ 53.54万
• 依托单位: Lehigh University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-05-01 至 2026-04-30
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2045976&HistoricalAwards=false
• 关键词: CAREER; Distributed; Protocols; Primitives; Optimized

The proliferation of the Internet enables the delivery of widely used services to millions of users every day. One of the biggest challenges in designing the system software that supports these services is engineering the components of the software infrastructure to scale to handle extremely large workloads. Remote Direct Memory Access (RDMA) is an important technology to improve achievable scalability. RDMA allows servers to access remote server memory without interrupting their execution resulting on more efficient communication and the ability to serve massive numbers of user requests. This project studies synchronization. Synchronization is a barrier to scalability as it entails significant additional computation in order to coordinate among multiple entities but is mandatory to ensure data integrity. In particular, synchronization typically requires two-sided communication where the remote machine responds to data access requests. This proposal investigates one-sided RDMA-optimized high-performance synchronization techniques and will demonstrate how they can be realized in easy-to-use abstractions to develop infrastructural system software.The outcomes of this project are techniques that allow both specialized and non-expert programmers to build services that can exploit RDMA to handle millions of user requests without sacrificing performance. By leveraging the infrastructural software delivered as a result of this project, developers can focus most of their work on the core business logic of the service. The principles and findings of the project contribute to a twofold goal. First, they form a solid foundation to instruct future generations of developers who yearn for delivering services that target society at large. Second, they abstract away the challenging details, allowing programmers to leverage the project's solutions regardless of their technical depth in the realm of RDMA and synchronization. This not only simplifies the implementation of large high-performance services, but also evens the playing field between those who have the resources to tackle complex problems, and those who do not. These research outcomes will be formalized in teaching material and tools for graduate and undergraduate courses and K-12 activities through a broad education plan designed to encourage students towards computer science study with a path towards graduate study and to broaden participation and increase diversity in computer science, especially for women.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

互联网的普及使得每天能够向数百万用户提供广泛使用的服务。设计支持这些服务的系统软件的最大挑战之一是设计软件基础架构的组件以扩展以处理超大工作负载。远程直接内存访问（RDMA）是提高可扩展性的重要技术。RDMA允许服务器访问远程服务器内存，而不会中断它们的执行，从而提高通信效率，并能够为大量用户请求提供服务。本项目研究同步。 同步是可扩展性的障碍，因为它需要大量的额外计算，以便在多个实体之间进行协调，但必须确保数据完整性。特别是，同步通常需要双边通信，其中远程机器响应数据访问请求。 该提案研究了单方面的RDMA优化的高性能同步技术，并将演示如何在易于使用的抽象中实现这些技术，以开发基础设施系统software.The成果，这个项目的技术，允许专业和非专家程序员构建服务，可以利用RDMA处理数百万用户的请求，而不牺牲性能。通过利用作为该项目结果交付的基础架构软件，开发人员可以将大部分工作集中在服务的核心业务逻辑上。该项目的原则和结论有助于实现双重目标。首先，它们形成了一个坚实的基础，指导未来几代渴望提供面向整个社会的服务的开发人员。其次，它们抽象出了具有挑战性的细节，允许程序员利用项目的解决方案，而不管他们在RDMA和同步领域的技术深度如何。这不仅简化了大型高性能服务的实现，而且在那些有资源解决复杂问题的人和那些没有资源的人之间建立了公平的竞争环境。 这些研究成果将通过广泛的教育计划正式纳入研究生和本科课程以及K-12活动的教材和工具中，该计划旨在鼓励学生通过研究生学习的途径进行计算机科学研究，并扩大参与并增加计算机科学的多样性。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查进行评估，被认为值得支持的搜索.

项目成果

Brief Announcement: Asymmetric Mutual Exclusion for RDMA

简短公告：RDMA 的非对称互斥

• 发表时间: 2022
• 期刊: Leibniz international proceedings in informatics
• 作者: Nelson-Slivon, Jacob and

Opportunities and Limitations of Hardware Timestamps in Concurrent Data Structures

并发数据结构中硬件时间戳的机会和局限性

• DOI: 10.1109/ipdps54959.2023.00068
• 发表时间: 2023
• 期刊: Proceedings IEEE International Parallel and Distributed Processing Symposium
• 作者: Grimes, Olivia;Nelson-Slivon, Jacob;Hassan, Ahmed;Palmieri, Roberto
• 通讯作者: Palmieri, Roberto

Bundling linked data structures for linearizable range queries

捆绑链接数据结构以进行线性化范围查询

• DOI: 10.1145/3503221.3508412
• 发表时间: 2022
• 期刊: The 27th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming (PPoPP
• 作者: Nelson-Slivon, Jacob;Hassan, Ahmed;Palmieri, Roberto
• 通讯作者: Palmieri, Roberto

Distributed Multi-writer Multi-reader Atomic Register with Optimistically Fast Read and Write

• DOI: 10.1145/3558481.3591086
• 发表时间: 2023-04
• 期刊: Proceedings of the 35th ACM Symposium on Parallelism in Algorithms and Architectures
• 作者: Lewis Tseng;Neo Zhou;Cole Dumas;Tigran Bantikyan;R. Palmieri

POSTER: OCToPus: Semantic-aware Concurrency Control for Blockchain Transactions

• DOI: 10.1145/3627535.3638494
• 发表时间: 2024-02
• 期刊: Proceedings of the 29th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming
• 作者: dePaul Miller;H. F. Korth;Roberto Palmieri