Implementation and evaluation of adaptive garbage collection
Eiko Takaoka, Yoshio Tanaka, Masakazu Nakanishi · Systems and Computers in Japan · 2000
Some garbage collection (GC) algorithms use a given threshold to control their behavior. Objects are classified according to their lifetimes, and the optimum threshold is used to efficiently collect unnecessary objects. Determining the approximate threshold value is crucial to such algorithms because the threshold value has a strong influence on the algorithm efficiency. Since memory usage depends on how intensively applications are being run, it is necessary to dynamically adjust the threshold. Demographic feedback-mediated tenuring (DFMT) is an algorithm that adjusts the threshold dynamically according to cell usage pattern. In this scheme, permanent objects are copied repeatedly in the space for the young generation. Our adaptive garbage collection (AGC) is an efficient garbage collection algorithm that solves this problem. AGC adjusts the threshold dynamically according to the fraction of cells used. We propose GC cost as a measure for quantitatively evaluating the efficiency of generation-dependent garbage collection with the copy scheme, and compare AGC and FGC (GC using DFMT). We show that AGC performs more efficiently for most applications. © 2000 Scripta Technica, Syst Comp Jpn, 31(14): 83–90, 2000