Rust Memory Allocators

Replacing the default system allocator is one of the lowest-effort, highest-impact optimizations available to any Rust project. On musl targets (Alpine/static builds), it's nearly mandatory — musl's default allocator causes a 7x slowdown. The cross-language fastest-dynamic-arrays research makes the case forcefully: switching allocators moves benchmarks more than switching languages does. A Vec<T> with mimalloc outperforms a Vec<T> with system malloc by 2–3× on push-heavy workloads, while C, Rust, and C++ vectors with the same allocator land within 10% of each other.

The two choices

Quick start

// In main.rs — that's it
#[global_allocator]
static GLOBAL: mimalloc::MiMalloc = mimalloc::MiMalloc;

Arena allocators

For hot paths where you can batch-deallocate, bumpalo (~2 ns bump allocation) and slotmap (generational-index arenas) avoid the global allocator entirely. These complement, rather than replace, a global allocator choice.

Why this matters more than container choice

The dynamic-array research established that the allocator dominates almost every other variable. Rust's Vec<T> benefits from this implicitly: Vec's reallocation path delegates to the global allocator's grow, which on jemalloc/mimalloc/tcmalloc can perform in-place expansion without copying. The container is the same; the allocator decides whether growth is O(N) or O(1). One #[global_allocator] line is, by a wide margin, the highest-leverage optimization in the language.