The procedure is similar to the one used for intermediate
outputs. Once a reduce tasks finishes, it contacts the server
providing a hash of the final output. The server save hashes
from multiple reducer replicas and decides which ones are
the trustworthy replicas. With this information, the server
updates the tracker and creates a .torrent for each one of
the reducer’s outputs. These .torrent files will be consumed
by the BitTorrent client placed at the server. Therefore, after
the first reduce task validation finishes, the server can start
downloading the final output from multiple volunteer nodes.