What are the performance considerations when running very large analytical queries in Athena?

Since Athena is built on Presto and processes queries in a distributed manner, performance for very large queries benefits greatly from using compressed columnar formats like Parquet or ORC, partitioning data appropriately, avoiding small numbers of very large files by compacting them, and limiting the columns selected, since Athena has to scan and process every byte of data included in a query regardless of how it will ultimately be filtered.

SELECT customer_id, order_total FROM sales_parquet WHERE year = '2026' AND month = '09';