For files, the client performs compression and sends the compressed data over the wire. The compressed data is written to disk in its compressed form. In certain circumstances (for example, small writes or writes to a middle of a file), the client might send uncompressed data to the server, and the server performs the compression. That said, when you think about compression performance, assume compression occurs on the client.
Note: For MapR-DB Tables, compression occurs on the server at the column family (CF) level. This ensures efficient compression in the data stored in the same range of a CF (most likely from multiple clients). MapR Streams data is compressed on the client.
Compression is set on a per file, table, stream, and directory basis. It's inherited from the parent directory directories (see hadoop mfs -setcompression).
For more information about compression, see Compression in the MapR user documentation.
Posting reference to a related document showing a working example of compression at work Interpreting compression ratios in hadoop fs command outputs.
Retrieving data ...