ISBN 9783843955768

978-3-8439-5576-8, Reihe Informatik

Thanh-Dang Diep
Supporting the Development of Nonblocking Data Structures on Distributed Memory Systems

138 Seiten, Dissertation Ludwig-Maximilians-Universität München (2025), Softcover, B5

Zusammenfassung / Abstract

Nonblocking concurrent data structures (DSs) are an essential part of parallel applications with producer-consumer patterns, where they can help improve fault tolerance and performance. Although there are many nonblocking concurrent DSs widely used in practice, most of them are designed for shared-memory machines (SMMs) and cannot be used on distributed-memory machines (DMMs). Existing approaches to bridge the gap between SMMs and DMMs are based on special hardware support and specific programming languages, or they focus on tailor-made nonblocking distributed DSs for a few specific applications and omit the potential for adapting a great wealth of existing nonblocking DSs for SMMs in the literature to DMMs, losing generality. Hence, the question of how to support the development of nonblocking concurrent DSs on DMMs in a general and portable manner arises. To address this question, this thesis develops a 3-part answer. Part one attempts to establish the foundations of nonblocking distributed DSs. Part two develops a practical dynamic global memory allocator. Part three aims to make the global memory manager even easier to use by providing a new unified programming interface generalized from various well-known safe memory reclamation schemes in the literature. By means of our 3-part answer, most of the well-known nonblocking concurrent DSs in the literature can be easily ported from SMMs to DMMs in a general and portable manner, opening up the opportunities for further studies on the applicability of the distributed versions of the nonblocking concurrent DSs to the parallel applications for DMMs.