Design a Time-Series Tick Store

Start with what a tick is. Every venue publishes a stream of tiny events: a trade printed at a price and size, a quote updating the best bid or offer, an order added or cancelled deep in the book. Each of those is a tick. A tick store keeps all of them, forever, so that later someone can ask what the market looked like at any instant in the past. It is the recorded memory of the market. The record is the product. A researcher testing whether a signal predicts returns, a backtest simulating a strategy across five years, a trader checking why a fill came in where it did: all of them read from this store. The design question is how to write that firehose down cheaply and read arbitrary slices of it back fast.

Two forces pull on the design, and naming them early makes every later choice follow. The first is ingest. A full US market-data feed, equities across every venue plus the options feed, peaks at tens of millions of messages a second. Over a single trading day that is hundreds of billions of ticks. The store has to absorb all of it without dropping a message. The second is the query pattern. The reads are not "give me one row." They are "give me every trade in this symbol between these two timestamps," or "compute the volume-weighted average price for these two hundred symbols over this month." Range scans by symbol and time, often over enormous spans. Those two forces, a write firehose and heavy range reads, point straight at columnar storage partitioned by symbol and time. That is where this design lives.

The Citadel angle

This is the research substrate the whole firm reasons from, and it is a different game from the hot path. The order book and the order gateway you designed earlier are microsecond systems where a lost nanosecond is a lost trade. The tick store sits off that path entirely. Nobody is quoting a market from it; it is where the evidence lives after the fact.