Optimal Choice of When to Garbage Collect

Abstract

We consider the ultimate limits of program-specific garbage collector (GC) performance for real programs. We first characterize the GC schedule optimization problem. Based on this characterization, we develop a linear-time dynamic programming solution that, given a program run and heap size, computes an optimal schedule of collections for a non-generational collector. Using an analysis of a heap object graph of the program, we compute a property of heap objects that we call their pre-birth time. This information enables us to extend the non-generational GC schedule problem to the generational GC case in a way that also admits a dynamic programming solution with cost quadratic in the length of the trace (number of objects allocated). This improves our previously reported approximately optimal result. We further extend the two-generation dynamic program to any number of generations, allowing other generalizations as well. Our experimental results for two generations on traces from Java programs of the DaCapo benchmark suite show that there is considerable promise to reduce garbage collection costs for some programs by developing program-specific collection policies. For a given space budget, optimal schedules often obtain modest but useful time savings, and for a given time budget, optimal schedules can obtain considerable space savings. CCS Concepts: • Software and its engineering → Runtime environments;

Learned garbage collection

“…Prior work by Jacek et al gave a dynamic programming algorithm to compute the optimal GC locations (in terms of reducing GC execution time) given an execution trace of a program on a particular input [12]. Later, Jacek et al showed that supervised learning techniques could be used to generalize these execution traces to other inputs [11].…”

“…For some programs, the location where GC is performed can impact performance [12]. For example, if a function creates many allocations that will soon have their reference count reach zero (triggering an automatic collection), garbage collection may waste time scanning these allocations.…”

“…We present a preliminary sketch and prototype of a garbage collection policy powered by reinforcement learning (RL). While learning approaches to GC are not new [11,12], existing approaches require pre-training on captured execution traces, and try to minimize the time spent in GC mechanisms. We target long-running programs that repeatedly execute a core loop, such as a web server or a database, and we attempt to optimize a custom reward function, such as request latency or video delay.…”