CSR: Small: Fault-Tolerant Distributed Software Transactional Memory: Theory, Protocols, and Java Package

CSR：小型：容错分布式软件事务内存：理论、协议和 Java 包

• 批准号: 1217385
• 负责人: Binoy Ravindran
• 金额: $ 40万
• 依托单位: Virginia Polytechnic Institute and State University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2012
• 资助国家: 美国
• 起止时间: 2012-09-01 至 2016-12-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1217385&HistoricalAwards=false
• 关键词: CSR; Small; Fault; Tolerant; Distributed

The project is to develop protocols, mechanisms, and a Java implementation of fault-tolerant distributed transactional memory (DTM). DTM promises to alleviate the programmability, scalability, and performance challenges of lock-based distributed concurrency control. Fault-tolerance is essential to DTM to cope with node/network failures, and object replication is central toward achieving this. Replication protocols must be scalable and ensure transactional correctness and progress properties. The expected outcome is a novel replicated DTM framework, whose key idea is to split transactions into two independent phases with orthogonal responsibilities: 1) regular read/write phases, which quickly look-up and fetch latest copies of required objects without concurrency control, and 2) a request-commit phase, which does (distributed) concurrency control. The project investigates the use of quorum-based replication protocols, which maintain transactional metadata in read/write quorums, and do scalable concurrency control by exploiting the quorum intersection property. The replicated DTM framework and protocols are implemented in the open-source HyFlow DTM Java package (hyflow.org). Fault tolerant DTM has potential to improve both reliability and performance of a broad range of advanced distributed computing applications, including defense systems. The project has plans to transitioning this technology (techniques and HyFlow implementation) to a production system of the US Department of Defense, to leverage the benefits of fault-tolerant DTM. This is a direct potential economic and social benefit of the research. Additionally, the project's results are being incorporated into a graduate course at Virginia Tech that includes students at Blacksburg, VA, scientists and engineers at US Naval Surface Warfare Center Dahlgren Division (NSWCDD), VA through Virginia Tech's graduate outreach program at NSWCDD, and students in the Middle East and North Africa through Virginia Tech's VT-MENA program at Egypt.

该项目旨在开发容错分布式事务存储器（DTM）的协议、机制和Java实现。DTM承诺减轻基于锁的分布式并发控制的可编程性、可伸缩性和性能挑战。容错对于DTM科普节点/网络故障是必不可少的，而对象复制是实现这一点的核心。复制协议必须是可伸缩的，并确保事务的正确性和进度属性。预期的结果是一个新的复制DTM框架，其关键思想是将事务分为两个独立的阶段，具有正交的责任：1）常规的读/写阶段，快速查找和获取所需对象的最新副本，而无需并发控制，以及2）请求提交阶段，进行（分布式）并发控制。该项目研究了基于仲裁的复制协议的使用，该协议在读/写仲裁中维护事务性元数据，并通过利用仲裁交叉点属性进行可扩展的并发控制。复制的DTM框架和协议在开源HyFlow DTM Java包（hyflow.org）中实现。容错DTM具有提高包括国防系统在内的广泛的高级分布式计算应用的可靠性和性能的潜力。 该项目计划将该技术（技术和HyFlow实施）转移到美国国防部的生产系统中，以利用容错DTM的优势。这是研究的直接潜在经济和社会效益。此外，该项目的结果正在纳入弗吉尼亚理工大学的研究生课程，其中包括弗吉尼亚州布莱克斯堡的学生，美国海军水面作战中心达尔格伦分部（NSWCDD）的科学家和工程师，弗吉尼亚理工大学通过NSWCDD的研究生推广计划，以及中东和北非的学生通过弗吉尼亚理工大学在埃及的VT-MENA计划。