Trading Systems & Market Connectivity
Szymon Kopyciński · 23 September 2026
1. Trading systems
Trading system
The software and infrastructure involved in making and executing trading decisions.
Trading stack
The collection of systems involved in the trading process.
A simplified stack:
Exchange market data
↓
Feed handler
↓
Order book
↓
Strategy
↓
Risk checks
↓
Order gateway
↓
Exchange
OMS
Order Management System.
Software that tracks the lifecycle and state of orders.
EMS
Execution Management System.
Software focused on executing trading decisions.
Many products combine OMS and EMS functions.
RMS
Risk Management System.
Software that monitors and enforces trading risk limits.
Order gateway
Software that sends orders to, and receives responses from, an exchange or broker.
Exchange gateway
The component handling communication with an exchange; similar to an order gateway.
Matching engine
The system within an exchange that matches compatible buy and sell orders.
Strategy engine
Software that runs trading logic and generates decisions.
Position service
A system that tracks current holdings or exposures.
Kill switch
A mechanism that rapidly stops trading activity, typically by cancelling outstanding orders and blocking new ones.
Pre-trade risk
Risk controls applied before an order is allowed to enter the market.
Examples include:
- maximum order size;
- maximum position;
- maximum notional exposure;
- price collars;
- trading permissions.
2. Trading protocols and networking
Protocol
A defined set of rules for communication between computer systems.
Exchanges publish protocols describing how participants send orders and receive market data.
FIX
Financial Information eXchange.
A widely used standard protocol for communicating orders, executions and other trading information.
Many exchanges also offer proprietary binary protocols for their lowest-latency interfaces.
Binary protocol
A protocol encoding messages in compact binary form rather than human-readable text.
Binary protocols use less bandwidth and are faster to parse than text protocols such as FIX.
TCP
Transmission Control Protocol.
A reliable, connection-based transport protocol.
TCP delivers a stream of bytes intact and in order, retransmitting lost data.
UDP
User Datagram Protocol.
A connectionless transport protocol with lower overhead than TCP and no guarantee of delivery or ordering.
Market-data feeds commonly use UDP because it supports multicast, and because a lost packet does not delay the packets behind it as it would under TCP.
Multicast
A networking mechanism that lets one sender distribute data to many receivers efficiently.
Exchange market-data feeds commonly use UDP multicast.
Packet
A unit of data transmitted across a network.
Market-data messages are carried inside network packets.
Packet loss
A network packet failing to reach its destination.
In market-data systems, packet loss creates sequence gaps.
3. Latency and performance
Latency
The delay between one event and another.
A trading system has many latencies, including:
- market-data latency;
- strategy latency;
- order-entry latency;
- exchange round-trip latency;
- end-to-end latency.
Round-trip time / RTT
The time required for a message to travel to another system and for a response to return.
Throughput
The amount of work a system can process over a period.
For example:
1 million market-data messages per second.
High throughput and low latency are related but distinct objectives.
Jitter
Variation in latency.
A system that usually responds in 10 microseconds but occasionally takes 10 milliseconds may have excellent average latency but terrible jitter.
Tail latency
Latency experienced by the slowest fraction of events.
In trading, slow outliers often coincide with busy markets, which is when speed matters most.
p50
The 50th percentile latency, or median latency.
Half of observations are faster and half are slower.
p99
The 99th percentile latency.
99% of observations are faster than this value; 1% are slower.
p99.9
The 99.9th percentile latency.
Used to track rare but damaging delays.
Microsecond / µs
One millionth of a second: $1,\mu\text{s} = 10^{-6},\text{s}$.
Nanosecond / ns
One billionth of a second: $1,\text{ns} = 10^{-9},\text{s}$.
Colocation
Placing trading servers in the same data centre as an exchange's matching engine.
Shorter physical distance reduces network latency.
Bare metal
A physical server dedicated to a single workload rather than a virtual machine sharing hardware with other customers.
Bare-metal infrastructure is common where predictable performance matters.
Kernel bypass
Techniques that let applications process network traffic without going through the operating system's normal networking path.
This can substantially reduce latency and jitter.
Clock synchronisation
Keeping clocks across different machines aligned.
Accurate timestamps are needed to analyse events across distributed trading systems. Common protocols are NTP (Network Time Protocol) and PTP (Precision Time Protocol); PTP achieves much tighter synchronisation.
Part of Glossary Index · Previous: Derivatives: Futures, Options & Greeks · Next: Production Engineering for Trading
NEFS Quant · About · Blog · News & Events · People · Privacy policy