2026/05/31 by Jake Yoon · 2 voices
Computer Science · #cs.DC #cs.DB #cs.DS #cs.PF
arxiv published 2026/05/31 · arxiv updated 2026/07/15
We drove 247 matching engines through one C-ABI harness on one identical workload: every open-source FIFO implementation we could find, deduplicated, and our own, on the same gate. The workload doubles as a byte-identical correctness oracle - 1,000,000,000+ order messages per engine, replayed against an independent-engine consensus. Only 47 are correct as shipped; we filed 181 GitHub issues upstream, 18 already fixed by their maintainers, none declined. Our engine leads the 160 that reproduce the consensus by ~25 M/s - 4x the second best - on worst-case throughput. One core sustains 33.2 million order messages per second at sub-microsecond median latency; a 96-core server (~1,630/month) sustains ~640 million/s across 10,000 symbols, over 20x the CTA quote feed's capacity. The lead is structural: the 52 engines written inside the trading industry sit under the same 8.19 M/s ceiling as the rest of the field. Every classical book - linked lists in a balanced tree - pays a pointer chase and a root-to-leaf search per operation. We eliminate both: the Priority-Indicated Node (PIN) resolves insertion in O(1) from priority indicators, not O(log n) comparisons; a neighbor-aware balanced tree splices or removes a price level in O(1) writes from the in-order neighbors electronic trading already supplies.