Allocator choice can matter when a FIX processing path creates many short-lived objects. We have not published reproducible measurement results or raw runs that support the previous mimalloc, jemalloc and tcmalloc latency ranking, so the percentage results and version-specific comparison have been withdrawn.
Comparable workloads
Test the same message corpus and allocation pattern with each allocator. Separate allocation-heavy parsing from a pre-allocated hot path, because the latter may spend little time in the allocator.
What to publish
Record exact allocator and compiler versions, CPU and operating system, build flags, thread count, warm-up, raw latency samples and P50/P99 calculations. Include the benchmark command and source commit. Without those artifacts, a winner or expected latency reduction cannot be established for another deployment.
