Garbage collecting the Internet: a survey of distributed garbage collection

Reviewer: Armin B. Cremers

Motivated by the advent of programming languages like Java, which are oriented to the needs of applications distributed over the Internet, this paper's objective is to give readers an overview of the range of solutions for the problem of identifying and reclaiming unreferenced distributed memory cells. Already a nontrivial problem in single-address-space systems, the distributed version of the problem raises the additional issues of communication overhead and nondeterministic latency. The authors emphasize the roots of the distributed solutions in the earlier single-address-space approaches by first developing a taxonomy of the latter according to the different techniques employed for identifying and reclaiming, and then structuring the primary content of the paper based on this taxonomy, which describes 15 approaches to the problem. The conclusion presents a carefully developed, qualitative comparison of the approaches discussed earlier. Finally, the design of distributed garbage collection is generally characterized as the delicate problem of balancing the different determining properties of such a scheme, with the improvement of one aspect always reducing the quality of another. As future work, the authors suggest a more systematic comparison of the numerous existing approaches. The paper is elegantly crafted, especially in the way the authors guide readers from the origins of the problem area to current work. The various approaches are described and related well, making the paper an excellent introduction to the topic.