If a fixed exponentially decreasing probability distribution function is used to model every object’s lifetime, then the age of an object gives no information about its future life expectancy. This radioactive decay model implies there can be no rational basis for deciding which live objects should be promoted to another generation. Yet there remains a rational basis for deciding how many objects to promote, when to collect garbage, and which generations to collect. Analysis of the model leads to a new kind of generational garbage collector whose effectiveness does not depend upon heuristics that predict which objects will live longer than others. This result provides insight into the computational advantages of generational garbage collection, with implications for the management of objects whose life expectancies are difficult to predict. 1

Generational collection has improved the efficiency of garbage collection in fast-allocating programs by focusing on collecting young garbage, but has done little to reduce the cost of collecting a heap containing large amounts of older data. A new generational technique, older-first collection, shows promise in its ability to manage older data.

implementations incorporate techniques that restore robustness by effectively disabling older-first collection in cases where degradation occurs, and this dissertation discusses further measures the collectors may take to avoid performance degradation. i I thank my committee members, Mitchell Wand, Rajmohan Rajaraman and Eliot Moss, and particularly my advisor, Will Clinger, for their participation, comments and help. Will Clinger also contributed in many ways over many years to Larceny, my experimental system, most obviously by writing the Twobit compiler for Scheme. The benchmarks used in this dissertation were written or modified by or in other ways contributed to by Henry Baker, Hans Boehm, Bob Boyer, Will Clinger, David Detlefs, John Ellis, Marc Feeley, Richard Gabriel, Pete Kovac, Eugene Luks, and Richard O'Keefe. The work was supported in part by NSF Grant CCR-9629801 and by the College of Computer Science at Northeastern University. The most profound thanks still go to m