Choosing a Set of Partitions to Collect in a Connectivity-Based Garbage Collector

Connectivity-Based Garbage Collection is a new family of garbage collection algorithms that are based on potential object connectivity properties. Objects are placed into partitions based on a static connectivity analysis. When the program needs memory, the collector can choose a set of partitions to do a garbage collection on. This choice should maximize the expected benefit in reclaimed memory, while minimizing the cost in expended work. We formalize this problem and present the flow-based chooser, an algorithm that uses network flow to find an optimal solution. We compare it to the greedy chooser, a simpler algorithm that may not find an optimal solution.