Market data receive
Receive exchange market data through a low-jitter network path with kernel bypass where required.
Reference architectures
Designing a production-grade low-latency trading platform.
A reference architecture for deterministic market data processing, strategy execution, risk controls, order management and reliable downstream distribution.
Domain: FinTech systemsFocus: Low latencyArchitecture: Event driven
Design objective
Keep the trading path deterministic.
The critical path should do only the work required to transform market events into validated exchange orders.
Storage, analytics, dashboards, reporting and research remain outside the latency-sensitive process and consume published events asynchronously.
01
Pinned CPU cores
02
Preallocated memory
03
No blocking calls
04
Lock-free communication
05
Explicit state ownership
06
Deterministic critical path
System architecture
Separate the deterministic trading path from the supporting platform.
Select or hover over a component to inspect its responsibilities, CPU placement and implementation choices.
Latency-sensitive processing
Exchange boundary
Supporting and downstream systems
Colocated trading system
Interactive architecture
Asynchronous event publishing
CPU and thread architecture
Explicit ownership, CPU affinity and predictable communication.
Critical threads are pinned to dedicated cores, kept on the same NUMA node and connected through single-producer, single-consumer queues.
Queue path
Receiver
Normalizer
Book Builder
Strategy
OMS
Risk
Gateway
Each boundary uses an SPSC ring buffer with cache-line separation between producer and consumer state.
NUMA 0
Latency-sensitive cores
NUMA 1
Supporting workloads
Reserved cores
Monitoring and logging
Metrics, health checks and asynchronous log flushing.
Admin and configuration
Authenticated risk and strategy configuration updates.
Critical path
From exchange packet to validated order.
Decode and normalize
Decode exchange-specific protocols and convert them into stable, exchange-agnostic events.
Order book and market state
Maintain order books, trades, reference data and strategy-facing market state.
Strategy execution
Run strategy logic against deterministic snapshots without blocking calls or runtime allocation.
Risk and order management
Apply pre-trade controls, position limits and order lifecycle management before transmission.
Exchange gateway
Encode and transmit orders while processing acknowledgements, fills, rejects and cancels.
Supporting platform
Everything important does not belong in the critical path.
Downstream publishing
Publish market data, orders, executions, positions and risk events outside the critical path.
State replication
Replicate strategy attribution, positions and execution state to a warm standby environment.
Historical data
Store normalized market data and execution events for replay, analytics and research.
Operations
Provide time synchronization, health monitoring, latency measurement, alerting and auditability.
Engineering partnership
Building latency-sensitive infrastructure?
Let's discuss your exchange connectivity, execution path, risk model and operational requirements.