Reconsidering "Reconsidering Custom Memory Allocation"

TL;DR

This paper reevaluates the effectiveness of custom memory allocators on modern hardware, confirming their locality benefits and extending prior findings with new benchmarks and methodologies.

Contribution

It revisits classic conclusions about custom allocators using modern hardware, benchmarks with real-world applications, and introduces a fragmentation-locality exploration methodology.

Findings

Region-based allocators improve locality and execution time.

Per-class allocators show no significant speedup.

Results support and extend original conclusions.

Abstract

Programmers using native languages such as C, C++, or Rust can implement custom memory allocation strategies to improve execution time. In their paper titled "Reconsidering Custom Memory Allocation" almost 25 years ago, Berger et al. showed that while per-class allocators provide no significant speedups over a state-of-the-art general-purpose allocator, region-based allocators can improve execution time by allocating and freeing objects in bulk. This paper revisits that work on a modern hardware platform with modern general-purpose allocators to evaluate whether their conclusions still hold. It also augments the benchmark suite with two large real-world applications (Clang and Blender), and introduces a methodology to explore the effect of memory fragmentation on locality in general-purpose allocators. Our results support and extend the original conclusions, demonstrating the locality advantages of region-based custom memory allocators.