Parquet vs JSON

Apache Parquet is an open-source binary columnar storage format. It stores data column by column, embeds the schema in the file footer, and applies efficient compression (Snappy or Zstandard by default). A JSON dataset converted to Parquet typically shrinks to 5–15% of its original size. Parquet is the native format of AWS Athena, Google BigQuery, Apache Spark, and most cloud data warehouse platforms.

Parquet's columnar layout enables a critical performance optimisation: a query that reads only three of a table's twenty columns scans roughly 15% of the file. On a dataset with billions of rows, this difference in I/O cost translates directly into query cost and latency. Parquet is the reason modern cloud data warehouses can run analytical queries cheaply at scale.

JSON (JavaScript Object Notation) is a plain-text format supporting objects, arrays, nested structures, and native data types. It is the standard format for REST APIs, NoSQL document databases (MongoDB, Firestore, DynamoDB), configuration files, and application event logs. JSON is human-readable and supported by every programming language and web platform.

JSON's flexibility is its key differentiator: a JSON document can have variable structure, deeply nested fields, and arrays of objects within objects. This makes JSON ideal for data that doesn't fit a rigid tabular schema. The tradeoff is verbosity — in a large JSON array, every record repeats every key name, making JSON far larger than equivalent columnar formats.