Scaling and benchmarking a critical message bus using a new indexing strategy
Summary
Jane Street's post describes scaling Aria, their internal messaging framework, by introducing topic-partition indexing and a shared block pool. The changes reduce CPU usage by about 30% in production workloads and lower latency for tip and initial recovery, validated through profiling, expect tests, and property-based testing, with adaptive subtree-store splitting deployed in staging and production soon.