Mandiri Sekuritas

Lately I’ve been building a small exchange simulation using OUCH (order entry), ITCH (market data), and ITCH-MDF (reference feeds). In an earlier version, I used mmap between the matching engine and ITCH. It was fast and efficient, but also introduced complexity—especially around synchronization, memory lifecycle, versioning, and recovery. In the current iteration, I switched to a simpler in-process pipeline using lock-free SPSC queues: OUCH clients -> SoupBinTCP gateway -> matching engine -> SPSC event queues -> ITCH publisher -> ITCH-MDF publisher The tradeoffs are clearer: - mmap is useful for low-overhead IPC - SPSC is easier to reason about (ownership, ordering, determinism) - but it only fits naturally in a single producer → single consumer flow The ASX Trade documentation (with 4 matching engines) highlights this well: the real challenge isn’t just speed, but keeping sequencing, replay, and feeds consistent across engines. When moving to multiple matching engines, SPSC alone isn’t enough. Need routing, partitioning (e.g. by instrument), per-engine queues, and aggregation on the publishing side. This project keeps reinforcing how much details matter—protocols, matching logic, queues, and recovery all shape the system. References: https://lnkd.in/g6Zyqzip https://lnkd.in/gRujK2Sx Still a simulator, but a useful way to explore real-world tradeoffs. #marketdata #capitalmarkets #hft #matchingengine #tradingsystem #itch #ouch