On measuring garbage collection responsiveness

In this article we survey and evaluate methods for measuring and/or illustrating the responsiveness of low-latency garbage collectors. These methods include pause time distributions, minimum and bounded mutator utilization curves, percentile utilization curves, and cathedral graphs; the latter we introduce. We also discuss why we believe it is important to evaluate a garbage collector on its compliance against an application-specific goal. We propose to do so with two techniques: Vmetrics and GC overhead graphs.