Trading-System

An asynchronous and low-latency trading system designed under service-oriented architecture (SOA).

The project gives an example of a trading system for seven US Treasury securities, with real-time price and orderbook flows, data streaming, user inquiries, algorithm order execution, risk monitor, logging modules, etc.

Service Architecture

Data flow into the trading system through Subscribe() that calls OnMessage(), transmits among different service components through ProcessAdd(), and flows out of the system through Publish().

Connector

Connectors can be subscribe-only or publish-only, or both. An input file connector can both Subscribe data from outside data sources and Publish data to a socket with a specified host and port. An inbound connector can then subscribe data from the socket and flow data into the trading system by calling Service.OnMessage(), or publish data to outside source using Service.Publish().

Client-Server Pattern

The whole system follows a client-server pattern through asynchronous I/O. The program consists of four client programs that subscribe to external data and publish to TCP sockets, and a main server program running six servers simultaneously using multi-threading. Data flows into the trading system through connectors from connectivity source (e.g. a socket, database, etc).

Socket-based Communication

Communication between different components is based on sockets to ensure real-time, low-latency and high throughput. Six servers are running simultaneously, with four (price, market, trade, inquiry server) listening to TCP sockets from localhost:3000 to localhost:3004 and flow data into the system, and two (steaming and execution server) publishing data to TCP sockets localhost:3004 and localhost:3005.

Deployment

The instruction to run the system on a Linux machine is as follows (tested on Ubuntu 22.04). Before running the system, make sure you have installed boost and cmake tools, and check firewall settings to open local ports 3000-3005 for TCP sockets (important!)

# install boost and cmake tools
sudo apt-get update
sudo apt-get install libboost-all-dev
sudo apt install cmake

The direct way is simply run the bash script run.sh in the root directory. It starts the system and the servers in the foreground, then start the four client programs in the background.

chmod +x run.sh
./run.sh

Or one can manually build and run the system.

# compile and start servers
mkdir build
cd build
cmake ..
make
./server

# start clients in separate terminals
./price
./market
./trade
./inquiry

Scripts

• Main program

  • InputPriceConnector: an input connector that subscribes external price data and publishes to TCP socket localhost:3000
  • InputMarketDataConnector: an input connector that subscribes external orderbook data and publishes to TCP socket localhost:3001
  • InputTradeConnector: an input connector that subscribes external trade data and publishes to TCP socket localhost:3002
  • InputInquiryConnector: an input connector that subscribes external user inquiry data and publishes to TCP socket localhost:3003
  • StreamOutputConnector: an outbound connector that subscribes streaming flow data from TCP socket localhost:3000 and publishes to localhost:3004
  • ExecutionOutputConnector: an outbound connector that subscribes execution data from TCP socket localhost:3001 and publishes to localhost:3005
  • main: connect different services, attach six threads to six servers, and start listening to TCP sockets localhost:3000-3005

• Service components

  • pricingservice: read in price data from the socket to the system through an inbound connector
  • algostreamingservice: listen to pricing service, flow in data of Price<T> and generate data of AlgoStream<T>
  • streamingservice: listen to algo streaming service, flow in data of AlgoStream<T> and record bid/ask prices into priceStream<T>, publish streams via socket in a separate process
  • guiservice: a GUI component that listens to streaming prices that should be throttled with a 300 millisecond throttle., register a service listener on the pricing service and output the updates with a timestamp with millisecond precision to a file gui.txt.
  • marketdataservice: read in orderbook data from the socket to the system through an inbound connector
  • algoexecutionservice: listen to market data service, flow in data of Orderbook<T> and turn into execution data AlgoExecution<T>
  • executionservice: listen to algo execution service, flow in data of AlgoExecution<T> and record order information into ExecutionOrder<T>, publish order executions via socket in a separate process
  • tradebookingservice: read in trade data, listen to execution service at the same time, flow in ExecutionOrder<T> and turn in trade data of type Trade<T>
  • positionservice: listen to trade booking service, flow in Trade<T> data and turn into Position<T>
  • riskservice: listen to position service, flow in Position<T> data and calculate corresponding position risks, such as PV01<T>.
  • inquiryservice: read in user inquiry data, interact with connectors and deal with inquiries

• Other components

  • products: define the class for the trading products, which can be treasury bonds, interest rate swaps, future, commodity, or any user-defined product object
  • historicaldataservice: a last-step service that listens to position service, risk service, execution service, streaming service, and inquiry service; persist objects it receives and saves the data into a database (usually data centers, KDB database, etc)
  • utils: time displayer, data generator, and risk calculator

• Data and results

  • data: data source for price data, orderbook updates, user inquiries, and trade data, can be replaced by other connectivity sources (a database, socket, etc)

  • res: results published by the system, including processed queries, executed orders, positions, risk monitor, data streaming, and GUI output.

Note

The trading system is designed to be scalable, extensible, and maintainable. Multi-threading and asynchronous programming ensure low-latency, high throughput, and high-performance. The system is also designed to be modularized, with each service component being independent and loosely coupled with others. New trading products can be added into products.hpp, new services, listeners, and connectors can all be easily added and integrated into the whole system.