Millions of Parquet files in S3 are just "byte blobs" without metadata. Without knowing columns, types, partition keys, you can't write a single SQL statement. AWS Glue Data Catalog centrally manages this "data about data" — metadata. Exam scenarios like "Athena queries S3," "Redshift Spectrum reads external tables," "EMR shares Hive metastore" almost always involve Glue Data Catalog.
Data Catalog is fundamentally a managed metadata store compatible with Apache Hive Metastore. If you've run Hive Metastore on RDS, Glue takes that operational burden.
Catalog has hierarchical structure.