Parquet-MR contains the java implementation of the Parquet format. Parquet is a columnar storage format for Hadoop; it provides efficient storage and encoding of data. Parquet uses the record shredding and assembly algorithm described in the Dremel paper to represent nested structures.