Matching engine rebuilt to handle €2.2M in daily derivatives volume
Cutting order execution latency from 15ms to under 400µs to win back institutional market makers.
Three of Volt's five largest market makers had left in eight months. The reason was simple: 15ms P99 latency made the platform unsuitable for algorithmic strategies. The Python-and-Redis order book had been pushed to its limits, and any rewrite risked introducing correctness bugs in live matching logic.
We rewrote the engine in Rust, porting Python logic test-case by test-case before touching performance. Lock-free ring buffers, a price-level skiplist order book, and cache-line-aligned nodes. The network and persistence layers stayed unchanged until the new core passed equivalence testing against 90 days of replayed production order flow.
P99 latency settled at 380µs on standard cloud hardware. Two of the three departed market makers returned within three months. Daily notional volume reached €2.3B by Q4, up from €520M at project start.

Have a Similar
Challenge?
Let's discuss how we can help you achieve similar results.
Start a Conversation