Learned garbage collection

Abstract

Several programming languages use garbage collectors (GCs) to automatically manage memory for the programmer. Such collectors must decide when to look for unreachable objects to free, which can have a large performance impact on some applications. In this preliminary work, we propose a design for a learned garbage collector that autonomously learns over time when to perform collections. By using reinforcement learning, our design can incorporate user-defined reward functions, allowing an autonomous garbage collector to learn to optimize the exact metric the user desires (e.g., request latency or queries per second). We conduct an initial experimental study on a prototype, demonstrating that an approach based on tabular Q learning may be promising.

A memory footprint optimization framework for Python applications targeting edge devices

“…A similar approach in [26] describes a dynamic memory leak detection method for dynamic languages. The work in [27] introduces a garbage collector, which is enabled by machine learning methods to allow efficient memory deallocation. The authors of [28] propose a hardware-software co-optimization memory management scheme for dynamic languages, including Python.…”

MP-CodeCheck: Evolving Logical Expression Code Anomaly Learning with Iterative Self-Supervision

“…Software debugging has been found to consume upwards of 50% of all software development time [14]. Machine programming (MP), the field concerned with the automation of all aspects of software development [21], has seen advances in software development related tasks such as code auto-completion [12,24,54], code generation and program synthesis [6,7,18,22,23,26,28,31,34,37,39,41,43,45,52,55,59,60], program transformation [20,36] and repair [4,8,16,32,57], code similarity and recommendation [35,58], learned optimizations [10,38,44], and performance regression testing [3,5,30,42,53], amongst others.…”

Analyzing a Caching Model

“…In recent years, modern machine learning (ML) techniques have seen increasing interest in the computer systems community (Maas, 2020), including black box techniques such as neural networks. Research has demonstrated that ML can be successfully applied to improve systems tasks such as caching (Song et al, 2020;Liu et al, 2020;Zhou and Maas, 2021), job scheduling (Mao et al, 2019), memory management Cen et al, 2020), and prefetching (Hashemi et al, 2018). The general strategy of these approaches is to replace a hard-coded system policy, such as a caching policy or a branch predictor, with a learned model.…”