Implementing a LineHandler for Fast, Reliable Trading
For high-frequency trading, speed is everything. We worked with our client to design and implement a high-throughput, low-latency LineHandler, capable of operating reliably at microsecond-level latency. Our solution achieved an average latency of 1.35 microseconds, comparing very favourably with international high-frequency trading benchmarks, where trading gateways are expected to achieve under 6 microseconds one-way latency from application to wire ( the time from when an order message enters the system until it is written to the network socket ).
The LineHandler was implemented as part of a Direct Market Access (DMA) system that takes advantage of ultra-low latency, co-located within the exchange infrastructure. Our target application supports high-frequency trading strategies such as algorithmic trading and smart order routing.
LineHandler is a key component of the overall solution, providing a high - performance, exchange - specific communication layer between brokers and stock exchanges.
- The solution supports both stocks and convertible bonds (CBs).
- It is designed to accommodate the specific protocols of target exchanges, in this case fully supporting Arrowhead 4.0 (for Tokyo and other Japanese stock exchanges) for handling orders and notices.
- It ensures seamless message handling: it can send 40,000 order messages per second (200 messages per second per virtual server across 200 virtual servers – the maximum limit allowed by the stock exchange), while exchanges send notices at nearly the same rate.
- The system demonstrated exceptional low-latency performance, with a minimum of 1 μs, maximum of 28 μs, average of 1.35 μs, and median of 1 μs for sending order messages.
- The solution integrates cleanly with broker systems, minimising integration effort and risk.
The system’s performance results from a combination of techniques: a deterministic, event-driven FSM that removes unnecessary branching; a custom low-latency, lock-free message queue optimised for high-volume throughput; and a multithreaded design with dedicated processing loops for parallelism. Critical threads are pinned to specific CPU cores, reducing scheduler jitter and context switching to maintain consistent microsecond-level timing. Pre-allocated memory pools eliminate runtime allocation, ensuring fully predictable execution.
The LineHandler behaviour was validated through pre-production testing in a TSE-simulated Arrowhead environment, conducting real-time message-exchange tests to ensure protocol-level compliance, message accuracy, and throughput. It has since been exercised in trial sessions over a live exchange connection, confirming operational readiness and stable performance under production-like conditions.
