With a dataset that small (1 billion rows of 20 columns of small data items per year), just use a regular RDBMS like Oracle or Postgres. A small cluster will be plenty of capacity.