GoKawiilJane Street intern cuts Aria message bus CPU use 30% with new indexing method
Jane Street intern Theodor Totev redesigned how the firm's internal Aria messaging system filters subscriber data during 'tip recovery,' the process where lagging clients pull recent messages from an in-memory ring buffer. By replacing a simple linear scan with an indexing and tree-splitting approach tailored to Aria's hierarchical topic structure, the project cut CPU usage by 30% on production workloads while preserving the system's strict correctness requirements.
GoKawiil's interpretation of the reporting above, not reported fact.
Aria handles multiple terabytes of data daily, so a 30% CPU reduction in a core filtering path could meaningfully lower infrastructure costs and improve headroom as client load grows. The work also illustrates how indexing structures can outperform cache-friendly linear scans once data volume and subscriber counts scale past a certain threshold, a tradeoff relevant to other high-throughput messaging systems.
- Jane Street intern Theodor Totev built a new indexing strategy for the Aria internal message bus.
- The optimization reduced CPU usage by 30% on production workloads during 'tip recovery' message filtering.
- The change addressed scaling pressure as more clients subscribed to subsets of Aria's terabyte-scale daily data stream.
Source: blog.janestreet.com, 2026-10-08
Published there as: “Scaling and benchmarking a critical message bus using a new indexing strategy”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.