HDFS Hadoop Distributed File System

An open implementation of the Google File System. A NameNode holds all metadata in memory; DataNodes store the blocks themselves (128 MB by default). Files are written once and read many times; there is no random write.

The design idea was to move computation to the data rather than the reverse, because the network was the bottleneck. The cloud inverted that premise — S3 with separated compute and storage has largely displaced HDFS for new systems.