Transactional Threads for Reliable Persistent Application Systems

在存储和操作持久数据的高质量软件组件的开发中，该项目解决了两个重要的高级问题——可靠性和性能，在这些组件中，数据可以代表合作或竞争的主体同时共享和访问。研究的目的是设计、实现和评估允许应用程序开发人员在不牺牲性能的情况下，将传统数据库系统的竞争并发模型（即事务）与传统编程语言的合作并发模型（即线程）组合在一起的技术。由此产生的融合，即所谓的事务线程，鼓励开发更可靠的应用程序，这些应用程序可以操作大量共享的结构化数据。该项目正在开发集成编译器/运行时和数据存储技术的规范和实现，使事务线程能够有效地执行。通过执行和度量基准事务性持久性程序，对这些技术进行了实验性评估。具体的研究成果包括为Modula-3和Java编程语言的持久扩展开发事务性线程原型。本研究的实验方法为实验计算机科学学生的培训、进一步的实验研究应用和可靠的持久应用系统的开发提供了丰富的环境。http://www.cs.purdue.edu/~hosking

This project addresses two important high-level issues -- reliability and performance -- in the development of quality software components that store and manipulate persistent data, where that data may be shared and accessed concurrently on behalf of principals that may be acting either cooperatively or competitively. The objective of the research is to design, implement and evaluate techniques that allow application developers to compose the competitive concurrency model of traditional database systems, namely transactions, with the cooperative concurrency model of traditional programming languages - namely, threads -- without sacrificing performance. The resulting fusion, so-called transactional threads, encourages development of more reliable applications that manipulate large amounts of shared, structured data. This project is developing specifications and implementations of integrated compiler/run-time and data storage techniques that enable transactional threads to perform efficiently. These techniques are being evaluated experimentally through execution and measurement of benchmark transactional persistent programs. Concrete research products include the development of prototypes of transactional threads for persistent extensions of the Modula-3 and Java programming languages. The experimental approach to this research offers a rich environment for training of experimental computer science students, for further experimental research applications, and for development of reliable persistent application systems. http://www.cs.purdue.edu/~hosking