SHF: Small: Collaborative Research: PEGASUS: ProgrEss GuAranteeS for Universal tranSactions

The University Of Central Florida Board Of Trustees, Orlando FL

Investigators

Abstract

Software scalability and reliability on multi-core systems is a crucial and urgent issue that needs to be addressed. Transactional processing is a promising programming model for developing parallel applications across many domains, including scientific and application software, that simplifies the development of correct concurrent software. This project will develop transactional techniques that overcome many current limitations, and it will explore the optimization of these techniques on systems with support for transactions. The intellectual merits are the development of novel techniques for constructing transactional data structures on systems with and without hardware support for transactions. The project will create a first approach for transactional processing with progress guarantees. The project's broader significance and importance are to make accessible to the industry a set of prototype data structures and programming techniques that furthers the reliability and performance of software on current and future multi-core systems. The project will also develop critical human resources in systems programming. The project will lead to a number of fundamental and practical outcomes. Transactions have been recognized as a promising alternative to lock-based systems. However, transactions as currently implemented in software or hardware have different drawbacks, with the absence of a progress guarantee being the most fundamental. To overcome this problem, this work further develops the notion of transactional data structures that are based on lock-free techniques, thereby guaranteeing progress. The project will also extend the applicability of commutable transactions to various data structure types including commonly used linked and contiguous memory data structures. The project harnesses transactional hardware to accelerate the execution of common cases and demonstrates the applicability by prototyping a set of important containers. Finally, this project will evaluate the outcome against