How Deribit’s Matching Engine Reduced Latency and Stabilized Tail Performance
Summary
The article describes Deribit’s migration to a new matching engine designed to reduce order-handling latency and make it more consistent. It compares six-day periods before and after broad adoption, reporting substantial reductions in median, 90th-percentile, and 99th-percentile latency. The largest improvement was in tail behavior: the prior system’s slowest responses varied sharply with activity, while the new system kept them within a narrower range.
The change separated order entry and market data into dedicated paths and restructured some risk checks. A load example from two busy days reports more than a billion events, a high request burst, and a short-lived backlog that cleared quickly. These figures illustrate performance under the measured conditions, but the article does not provide the full technical methodology; it points to a separate paper for that detail. The new binary interfaces target latency-sensitive participants, while the Classic API remains supported.
Key ideas
- Latency percentiles reveal slow responses that an average or median can hide.
- The reported post-migration results show lower median and tail latency during the comparison period.
- Dedicated processing paths and fewer steps were used to improve order handling consistency.
- A busy-period example indicates that brief congestion occurred but cleared quickly.
- Performance figures describe specific measurement windows and should not be treated as universal guarantees.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.