I am trying to run FlinkSQL on a ton of historical batch data, say in Parquet files. What are some recommended configurations to get best performance here?