Trading Platform Shared Memory Matching Engine Latency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional trading platforms face challenges in performance, scalability, reliability, and efficiency, particularly in areas of latency, throughput, customer interfaces, testing, monitoring, and systems security, while also requiring reduced power and cooling demands and rapid scalability, resiliency, and fault isolation.
Innovation Solution
A hardware and software architecture that combines to provide ultra-low latency and scalable throughput, featuring a distributed processing system with multiple independent trading environments, shared memory for quote processing, and a multicast protocol for messaging, along with redundant services and a container concept for hardware scalability, to support high resiliency and narrow fault domains.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional trading platforms are used, then basic trading functions are provided, but performance and scalability are insufficient
Solution Approach 1:
The trading platform is divided into multiple independent trading environments (clouds), each handling specific symbols or instruments. This segmentation allows parallel processing across multiple environments, significantly increasing throughput while maintaining consistent performance through independent operation of each environment.
Solution Approach 2:
The system transitions from a single-threaded sequential processing model to a multi-dimensional parallel processing architecture using shared memory regions and multiple trading environments. This dimensional change enables simultaneous quote processing across multiple environments, dramatically improving throughput while maintaining reliability through independent failure isolation.
2Productivity
If more servers are added to increase capacity, then throughput improves, but power and cooling demands increase
Solution Approach 1:
Multiple trading environments share common infrastructure including hardware resources, operating system, and network connectivity through shared memory regions. This merging allows multiple processing environments to run on fewer physical servers, increasing processing capacity while reducing power and cooling demands compared to fully distributed architectures.
Solution Approach 2:
The shared memory regions and infrastructure serve multiple trading environments simultaneously, allowing a single server to perform multiple trading functions. This multi-functionality increases processing capacity without proportionally increasing power and cooling requirements.
3Ease of operation
If a centralized architecture is used, then system operation is simplified, but fault isolation becomes difficult
Solution Approach 1:
The system is segmented into independent trading environments that can be independently managed and monitored. Each environment handles specific symbols or instruments, providing natural fault isolation boundaries while maintaining simplified operation through standardized environment configurations and centralized monitoring capabilities.
Solution Approach 2:
Shared memory regions act as intermediaries between independent trading environments and the matching engine. This intermediary layer provides standardized communication protocols and data access patterns, simplifying system operation while maintaining fault isolation through the modular environment architecture.
4Loss of time
If custom hardware is designed for high performance, then latency is reduced, but device complexity increases
Solution Approach 1:
The system replaces complex custom hardware designs with software-optimized processing in standard server environments. Shared memory regions and direct memory access techniques eliminate mechanical I/O bottlenecks, achieving ultra-low latency through software architecture rather than specialized hardware, thereby reducing device complexity.
Solution Approach 2:
Shared memory regions serve as intermediaries that provide direct, high-speed access between trading environment applications and the matching engine without traditional I/O overhead. This intermediary mechanism achieves low latency through memory-level communication rather than complex hardware interfaces.
Data Source
AI summary
An electronic trading platform performs trading of one or more types of financial instruments and includes at least one cloud providing at least one independent trading environment executed by a server. The at least one cloud includes a plurality of instances of an express interface application executed by the server and configured to receive interests in the financial instruments. The plurality of instances of the express interface application are configured to write the interests to a shared memory on the server. A matching engine, executed by the server, is configured to continuously poll the shared memory, read the interests written by the plurality of instances of the express interface application, analyze the read interests to determine if there is a matching contraside interest and allocate matching interests based on at least one allocation rule.