Brokk AI Drops musl libc After Benchmarks Reveal 26% Performance Hit

Jonathan Ellis of Brokk AI has reported significant performance regressions after switching the Bifrost project to musl libc, leading the company to remove musl as a prebuilt option. The move follows a series of benchmarks revealing that the lightweight C library fails to match the efficiency of the industry-standard GNU C library (glibc).

Testing conducted on 4-core EC2 VMs showed that musl's default allocator is severely suboptimal. According to Ellis, "musl's allocator is indeed bad. Terrible, even." In an attempt to mitigate these losses, Brokk AI integrated mimalloc, a high-performance third-party allocator. However, the results remained disappointing; musl paired with mimalloc was still 26% slower than glibc in the tested scenarios.

The performance degradation was not limited to high-concurrency environments, as the hit was clearly observed on VMs using sized thread pools. Furthermore, the investigation identified that several common memory routines within musl, beyond just the allocator, were particularly slow.

The Trade-off of Simplicity

Developers, particularly those using Rust or C++, frequently adopt musl to create static binaries. This approach is popular for containerization—most notably within Alpine Linux—because it eliminates the complexities of dynamic linking and the risk of libc version mismatches in Docker images. By producing a single binary, developers can simplify deployment pipelines and reduce the footprint of their container images.

Implications for Runtime Efficiency

These findings highlight a critical trade-off: the convenience of "single binary" deployments may come at the cost of massive runtime inefficiency. For high-performance applications, the assumption that a third-party allocator like mimalloc or jemalloc can bridge the gap between musl and glibc is challenged by these results. The data suggests that the performance overhead is baked into the library's core memory routines, meaning that simply swapping the allocator is insufficient to achieve parity with glibc.

Looking Ahead

As Brokk AI reverts Bifrost to glibc to reclaim lost performance, the results serve as a warning to other developers prioritizing deployment simplicity over execution speed. While musl remains a staple for lightweight containers, these benchmarks suggest that for compute-intensive workloads, the standard GNU C library remains the necessary choice. It remains to be seen if future updates to musl's core memory routines will address these specific bottlenecks.