Garbage collection in a large LISP system

Abstract

This paper discusses garbage collection techniques used in a high-performance Lisp implementation with a large virtual memory, the Symbolics 3600. Particular attention is paid to practical issues and experience. In a large system problems of scale appear and the most straightforward garbagecollection techniques do not work well. Many of these problems involve the interaction of the garbage collector with demand-paged virtual memory. Some of the solutions adopted in the 3600 are presented, including incremental copying garbage collection, approximately depth-first copying, ephemeral objects, tagged architecture, and hardware assists. We discuss techniques for improving the efficiency of garbage collection by recognizing that objects in the Lisp world have a variety of lifetimes.The importance of designing the architecture and the hardware to facilitate garbage collection is stressed.reclaims their storage. The user program does not have to say explicitly "I'm done with this object." Automatic storage reclamation is usually called garbage collection. It can be thought of as finding all the objects that are no longer useful for anything-the garbage--and collecting the memory used to represent them so that it can be reused for new objects.Garbage collection consists of -Deciding when to garbage collect and how much of the machine's resources to devote to garbage collection (as opposed to "useful work").-Discovering the division of storage between garbage and good objects' representations.-Separating the garbage so that its memory can be reused without disturbing the good objects.

Using generational garbage collection to implement cache-conscious data placement

“…Moon describes a modification to this algorithm that results in approximate depth-first copying [18]. Wilson et al further refine the traversal to obtain hierarchical grouping of objects in TO space [30].…”

“…Researchers investigated two approaches to using a garbage collector to improve paging behavior of Smalltalk and LISP systems [18,30,15,10]. Static regrouping [18,30] uses the topology of heap data structures to rearrange structurally-related objects, while dynamic regrouping [10] clusters objects according to a program's data access pattern. Moon [18] found that depth-first copying generally yields better virtual memory performance than breadth-first copying for LISP, because it is more likely to place parents and offspring on the same page, particularly if data structures tend to be shallow, but wide.…”

“…Other researchers have attempted to improve a program's virtual memory (page) locality by changing the traversal algorithm used by a copying garbage collector [18,30,15,10]. We compare our cache-conscious copying scheme against one such algorithm (the Wilson-Lam-Moher algorithm [30]).…”

Recovery using virtual memory

“…In general, realtime garbage techniques are not suitable for such a machine. Real-time garbage collection techniques need special purpose hardware such as that found on the Lisp Machine [20] in order to be feasible. Thus, this thesis limits itself to a discussion of "stop the world" techniques.…”

“…The copying algorithm as presented in this chapter traverses and copies the graph of accessible objects in a breadth-first fashion. Moon [20] argues that, for Lisp, depth-first copying yields better locality than breadth-first copying. He suggests a method of copying called approximate depth-first copying, which uses depth-first copying on partially filled pages of to-space.…”