Use case - Building an FX high-frequency trading system

A company needs an HFT system capable of sending an order within 20 microseconds. To do this, the company can follow this approach:

• Choose a multi-process architecture over a multi-core architecture.
• Ensure each process is pinned to a specific core to reduce context switches.
• Have processes communicating over a circular buffer (lock-free data structure) in shared memory.
• Design the network stack using Solarflare OpenOnload for network acceleration.
• Increase the page size to reduce the number of TLB cache misses.
• Disable hyperthreading to get more control over the concurrency execution of the processes.
• Use the CRTP to reduce the number of virtual functions.
• Remove runtime decisions by using templated data structures.
• Pre-allocate data structures to avoid any allocation on the critical path.
• Send fake orders to keep caches hot and allow an order to go out at the last moment.

In...