A Rust service measures musl at 26 percent slower than glibc after switching allocators

Sources: primary, discussion
Summary: Jonathan Ellis measured Brokk's Bifrost service on 4-core EC2 VMs after moving Rust projects to musl for self-contained binaries, and reports that swapping in mimalloc recovers only part of the gap and leaves musl 26 percent slower than glibc. He states the allocator is not the whole story, because a workload that barely allocates regressed proportionally more than an allocation-heavy one, which he attributes to slow memory routines elsewhere in musl. Brokk is removing musl as a prebuilt option for Bifrost and keeping it only where a 25 percent penalty does not matter, on one team's measurement of two workloads at one instance size.
Why it matters: A static musl binary is a common default for container images, and this measurement puts a stated cost on that default for a service under load.