Assessing and Improving Garbage Collection Performance in the Julia Programming Language

Abstract

With the increasing popularity of the Julia programming language for memory-intensive applications, garbage collection (GC) is becoming a performance bottleneck, with reports of poor GC performance ranging from differential equation solvers to large database benchmarks.

There have been several GC optimizations (such as the implementation of a generational collector) targeting the Julia GC over the last decade, but none of them was in the direction of a multithreaded GC.

This thesis assesses GC performance in the Julia programming language and implements optimizations focusing on parallelizing automatic memory management routines.