An S3 analytics workload scans terabytes of JSON to answer queries for one date and a few columns. Which two changes most directly reduce scanned bytes? Select 2 answers.

1. Put a unique random value in every query cache key.
2. Convert suitable datasets to a columnar format such as Parquet.
3. Increase the public DNS TTL for the S3 bucket.
4. Partition data on useful query predicates such as event date.
5. Combine all history into one larger uncompressed JSON object.