The Case for a Read Barrier

Abstract

This paper looks at the performance of two different garbage collection algorithms on a very large and long running Lisp application. Both algorithms use write barriers for generational collection. Only one algorithm uses a read barrier for incremental collection. The results show little difference in the two algorithm�s ability to collect garbage and some difference in memory size. Any differences in CP[J usage were too small to be visible with the measuring techniques used. However, there were major differences in paging behavior when a read barrier permitted the garbage collector to work with the virtual memory manager instead of independently. © 1991, ACM. All rights reserved.