CSR:Small: Scalable Transaction Management and Consistency Models for Cloud Data Storage Systems

该项目的目标是为云计算系统中的事务管理开发可扩展的技术和系统架构。近年来，各种基于键值存储模型的数据管理系统应运而生，以满足基于云的应用程序的可扩展性要求。然而，这些系统不提供许多应用程序域中经常需要的丰富的数据管理原语，例如多行可序列化事务。该项目正在研究基于键值的数据存储系统的可伸缩事务管理技术，目的是支持此类系统中具有ACID(原子性、隔离性、持久性、一致性)属性的多行事务。对于可伸缩性，此调查基于使用快照隔离模型。在支持跨不同云数据中心的地理复制数据的事务管理方面也出现了其他问题。考虑到较弱的一致性模型可能适用于某些应用程序，该项目旨在开发支持不同事务一致性模型的技术，如快照隔离、因果一致性、可串行化以及云应用程序中地理复制数据的会话级别一致性。此外，该项目的目标是利用数据操作的语义来提高性能和可用性。该项目旨在支持将基于键值的云数据存储模型用于具有各种不同数据一致性要求的大类应用程序。云计算系统中缺乏具有强一致性模型的事务支持，这一直是此类系统在许多关键应用领域采用的障碍。该项目的一个潜在变革性影响将是使更多类别的应用程序能够利用云计算系统。

The goal of this project is to develop scalable techniques and system architectures for transaction management in cloud computing systems. In the recent years, various data management systems based on key-value storage models have emerged to address the scalability requirements of cloud based applications. However, these systems do not provide rich data management primitives, such as multi-row serializable transactions, often needed in many application domains. This project is investigating scalable transaction management techniques for key-value based data storage systems with the aim of supporting multi-row transactions with ACID (atomicity, isolation, durability, consistency) properties in such systems. For scalability, this investigation is based on utilizing the snapshot isolation model. Additional issues arise in supporting transaction management for geographically replicated data across different cloud datacenters. Considering that weaker consistency models can be adequate in certain applications, this project aims to develop techniques for supporting different transactional consistency models such as snapshot isolation, causal consistency, serializability, and session-level consistency for geo-replicated data in cloud applications. Moreover, the goal of this project is to exploit the semantics of data operations for enhanced performance and availability. This project aims at enabling the use of key-value based cloud data storage models for a large-class of applications with various different data consistency requirements. The lack of transaction support with strong consistency models in cloud computing systems has been an obstacle in adoption of such systems in many critical application domains. A potentially transformative impact of this project will be towards enabling a broader class of applications to utilize cloud computing systems.