On comprehensive global garbage Detection

Abstract

The experience gained with centralised garbage collection (GC) techniques has left a legacy of assumptions, expectations and tradeoffs, which may lead one to overlook some Global Garbage Detection (GGD) approaches. We argue that it is not necessary to give up on comprehensiveness in order to achieve a high degree of concurrency and scalability, although this may be at a price which is not palatable in a centralised system. For instance, much higher detection latency or space overhead, at least in terms of worst case scenarios, is acceptable in a large distributed system. We are currently implementing a variation of Schelvis' algorithm on Amadeus (taking advantage of Amadeus' object clustering ability), to conduct an empirical evaluation of the actual impact of such tradeoffs.