Research Initiation Award: Automatic Storage Management for Real-Time, Near-real-time, and Distributed Systems

本项目探索自动存储回收技术，特别是实时和分布式系统。本文从分布式数据的效率、实时保证和性能方面比较了增量垃圾收集技术。重点是在标准硬件和标准编译器优化上支持高性能的技术。项目主要分为三个阶段：(1)实现和评估了一种新的实时单处理器垃圾收集算法，该算法提供了比增量复制更好的实时性保证，因为应用程序和垃圾收集器之间的耦合较弱；(2)将算法扩展到近实时系统，在这种系统中，可以牺牲一些实时性保证来换取更高的性能；(3)在分布式垃圾收集系统中应用这些增量技术作为本地垃圾收集器，它们将对分布不佳的数据进行更高性能的垃圾收集。

This project explores techniques for automatic storage reclamation, especially for real-time and distributed systems. It compares techniques for incremental garbage collection in terms of efficiency, real-time guarantees, and performance for distributed data. The focus is on techniques that support high-performance on standard hardware, and standard compiler optimizations. The three major phases of the project are; (1) implementation and evaluation of a new real-time uniprocessor garbage collection algorithm which provides better real-time guarantees than incremental copying due to weaker coupling between the application and the garbage collector, (2) extensions of the algorithm for near real-time systems where some real-time guarantees can be sacrificed for higher performance, and (3) application of these incremental techniques as local garbage collectors in a distributed garbage collection system, where they will enable higher performance garbage collection of poorly distributed data.