This project is aimed to provide a safe, efficient and fast order matching system without any compromise in fairness which is very important in OMS software implementations.
By being safe, I mean the project must persist every trade crash-safely (not just provide an in-memory matching engine). Simultaneously it must maintain the fairness of transactions and the transparency of transaction execution. As a customer, Did you ever doubt about fairness of your buy/sell orders in a crypto exchange platform you use?!
By being efficient, I mean the project should use resources efficiently and reach reasonable benchmark numbers in peak loads. Using resources efficiently requires some custom implementations, like low level memory access and managing memory manually, thereby in implementation, I tried to use these kinds of techniques where ever possible. As an exchange platform founder, Did you ever been sensitized about how much efficient your main software components work?! Does your platform works well under heavy loads?! With how much cost and resource usage?
By being fast, I mean the processing of incoming orders, matching orders, persistence and fetching order books should all be fast and no bottleneck is acceptable. Why? Because you should compete with other platforms out there; Are your customers happy with you?! Or they complain about your platform's being slow sometimes?!
Pure java is used to provide a high performance asynchronous and non-blocking TCP level implementation for network stack. Also, a custom binary protocol is designed for messaging, and the network implementation directly uses that protocol to parse them.
A custom storage implementation is used in both asynchronous and synchronous manners to provide a solid context for working with files and file storage, including an atomic WAL implementation.
The project supports both one-core per symbol or one-core per multiple symbols architecture. The GC pressure is tried to be minimal, for example off-heap memory with manual control over allocation and de-allocation is nearly used all over the packages to gain better performance numbers.
The project is able to match 10000 crash-safe persisted orders per symbol (ex. BTC/USDT) per second. For example if you have two active symbols, on each symbol you easily can reach 10K matches, meaning 20K total. One-core per symbol or one-core per multiple symbols may affect these numbers.
- Price/Time algorithm
- Limit & Market orders
- Order canceling
- Order book fetching with optional depth parameter
- Remaining recording, at both order and trade models
- JFR enabled to be able to visualize detailed metrics about project's behavior at runtime
- FillOrKill order
- Partial cancelling
- Stop & StopLimit orders
- Restful API server
If you have any question about the project, or want to try OMS in a sandbox, or use in production, you can reach us at contact[at]openex[dot]software.