Therefore, we can leverage the task replication mechanisms
to share the burden of the data server. Even if the server
is unable to respond, a new mapper may continue to download
its input data from other mappers. The transfer bandwidth will also be bigger since a mapper may download input
data from multiple sources.