Ridgefield operates a large sensor fleet. It needs low-latency key-based reads and writes over time-series data, with no relational joins. Which data-store choice and design are most suitable?

1. A single relational VM enlarged indefinitely without access-pattern analysis.
2. Cloud Storage Archive with one object per individual database operation.
3. Bigtable with row keys designed for balanced writes and required reads.
4. BigQuery used for every synchronous per-sensor update.