Imagine you are managing a network of 1000 sensors, each generating time series data at the rate of one metric per sensor per second, accompanied by a timestamp. Currently, you have accumulated 1 TB of data and anticipate an increase of 1 GB daily. You require two distinct access patterns for this data:

1. Retrieve the metric from one specific sensor at a particular timestamp with a median latency in the single-digit milliseconds.
2. Perform complex analytical queries on the data, including joins, on a daily basis.

What would be the optimal way to store this data to meet these requirements?