HDFS is designed to solve the
limitations of NFS. It can store large amounts of data. It is fault tolerant (if a machine in the
cluster fails, it causes no disruption of service). If too many clients try to access the data resulting in a performance bottleneck, adding more nodes to the cluster remedy it. And finally, HDFS integrates well with
Hadoop MapReduce.
HDFS is modelled after GFS (Google Files System).
Weaknesses
Inefficient for small files (but the goal was to handle big files anyway).
Lack for (transparent) compression.
No random access to files: data can only appended to files.