A Boundary-First Benchmark for Low-Latency Crypto Trading Systems

Title: A Precise Benchmark for Low-Latency Crypto Trading Systems

A new benchmark for low-latency crypto trading systems has been introduced, focusing on the boundaries of the measurement process. The study performed on a laptop-class Intel Core i7-1360P processor shows that the run-level median latency is between 121.1-135.4 µs, while the run-level p99 latency is between 434.8-662.0 µs.

The measured path includes the journey of a synthetic top-of-book event to a locally generated order report, all within a single host. The process consists of depth, strategy, order input, TD (mock), and order report. The study excludes network sockets, exchange gateways, venue protocol encoding and decoding, matching engine time, acknowledgments, fills, queue position, market impact, and profitability from the measurement.

The study isolates three processes: market data, strategy, and trade execution, all communicating through a journal-based shared-memory event path. The journal serves as both the transport and evidence during the run, enabling process isolation and eliminating the need for a separate instrumentation layer. The strategy is a native C++ implementation, and the framework allows Python strategies without affecting the measured callback.

The study conducted five independent runs, each with 5,000 synthetic depth events at a 300,000 ns interval, and 1,000 orders emitted. The median latency is stable across runs, while the p99 latency varies between 434.8-662.0 µs. The stage latency breakdown shows that the depth-to-order input stage dominates the median latency, while both stages contribute to the tail latency.