Garbage Collection Evaluation Infrastructure for the Cog VM

Sophie Kaleba (1) , Clément Béra (2) , Eliot Miranda (3)

1 Université de Lille, Sciences et Technologies
2 VUB - Vrije Universiteit Brussel
3 Feenk

One of the next steps to improve Cog, the default virtual machine for multiple programming languages in the Smalltalk family, such as Pharo, Squeak and Newspeak, is to decrease garbage collection pause times. Reference garbage collection algorithm implementations and a benchmarking infrastructure are required to evaluate the performance of a new algorithm and compare it. Cog features a Mark-Compact algorithm, used in production, to which we added a Mark-Sweep algorithm, providing two reference algorithms. Benchmarks are built using two different approaches. Firstly, we turned code from memory intensive deployed applications into benchmarks to simulate real-world applications. Secondly, we built a configurable benchmark which simulates an application with different heap properties to be able to stress specific aspects of the memory management. We then evaluated the two reference algorithms on the infrastructure built to obtain reference benchmark results.

Mots clés: Benchmark; Garbage Collector; Virtual Machine; Managed Runtime

Domaines: Informatique [cs]; Informatique et langage [cs.CL]

Sophie Kaleba, Clément Béra, Eliot Miranda. Garbage Collection Evaluation Infrastructure for the Cog VM. International Workshop on Smalltalk Technologies, Sep 2018, Cagliari, Italy. ⟨hal-01874947⟩