On the Effects of Integrating Region-Based Memory Management and Generational Garbage Collection in ML

Abstract

We present a region-based memory management scheme with support for generational garbage collection. The scheme is implemented in the MLKit Standard ML compiler, which features a compile-time region inference algorithm. The compiler generates native x64 machine code and deploys region types at runtime to avoid write barrier problems and to support partly tag-free garbage collection. We measure the characteristics of the scheme, for a number of benchmarks, and compare it to the Mlton state-of-the-art Standard ML compiler and configurations of the MLKit with and without region inference and generational garbage collection enabled. Although region inference often serves the purpose of generations, we demonstrate that, in some cases, generational garbage collection combined with region inference is beneficial.