Dissecting Optiver’s HFT Optimization Playbook

Carl Cook from Optiver delivered what became one of CppCon’s most-watched talks on low-latency trading systems. Seven years later, some of his techniques remain foundational while others have aged poorly. Let’s examine what still holds water and what deserves skepticism.

The Speed Game: Context and Stakes

Cook frames the challenge perfectly: market makers need to update prices faster than competitors or risk getting picked off on stale quotes. The numbers he throws around - achieving 2.5 microsecond wire-to-wire latency - sound impressive until you realize modern systems are pushing sub-microsecond boundaries. But his core message stands: in electronic trading, being second means losing money.

The hotpath he describes - receiving market data, running strategy logic, executing risk checks, and firing orders - represents maybe 1% of the codebase but determines competitive survival. This code executes incredibly rarely, perhaps 0.01% of runtime, yet when it fires, every nanosecond counts. The entire computing stack fights against this pattern: operating systems want fair scheduling, networks batch packets for efficiency, and CPUs optimize for throughput over latency.