Latency doesn’t come from one place. It accumulates across the entire system

Lately I’ve been trying to understand how machines actually communicate and where latency comes from.

A few things that changed how I think:
- A socket is basically a communication endpoint between processes.
- IP addresses identify machines, but ports decide which process receives the data.
- TCP prioritizes reliability and correctness.
- UDP prioritizes speed and low latency.
- In HFT systems, stale market data can be more dangerous than missing data.

The biggest realization was that latency is everywhere.
Not just in slow internet, but in:
- network travel
- kernel processing
- memory access
- scheduling
- queues
- application logic

The deeper I go, the more I realize modern software is mostly systems exchanging messages under constraints.
And most optimization work is really about removing waiting.