A live trading system constantly reads and writes state — the current quote, open positions, cooldowns, recent signals. If it hit a spinning disk or a slow database for each of those, the trading loop would crawl. Fast systems keep this hot, frequently-changing state in memory — in RAM, or a dedicated in-memory data store — where access is measured in microseconds.
Why disk is too slow
Disk and traditional databases are built for durability, not speed — a round trip can take milliseconds, an eternity when your loop runs many times a second. For the working state that changes constantly and must be read instantly, in-memory access is orders of magnitude faster. The persistent database still has its place (the trade journal, config), but not in the hot path.
What in-memory stores solve
An in-memory store gives the trading loop a shared, blazing-fast scratchpad: the latest quotes any component can read, position state that updates atomically, transient flags that expire on their own. It's the system's short-term memory — fast enough to keep up with the tape, structured enough that every part of the system sees a consistent picture.
Durability lives on disk. Speed lives in RAM. A trading loop that confuses the two is a trading loop that's late.
Why it matters to you
This is the plumbing behind responsiveness. A system that keeps its live state in memory can react to a new tick without a storage bottleneck — the architectural reason it can act on a short-dated setup in real time. Pair it with an event loop and you have the foundation a tool like NoVo stands on: fast memory, fast reactions, honest state.