A PeerToPeer file downloading system which works per-file, unlike most others. Users first download a tiny .torrent file which provides MetaData on the download, including the URL of a centralised tracker that listens to special queries over an ordinary HTTP connection. The tracker itself does not have the file, it only keeps track of peers currently downloading the file to allow them to discover each other. The peers fetch sections of the file from each other in random order using a tit-for-tat system. The end result is a generally very efficient downloading system, offering insanely fast downloads even when the original provider of the file is on a very slow link.
To make the system work, a client which already has the full file, referred to as the seeder, must initially connect to the tracker. This client is in no way special and its continued presence is not required either. There needn't even be a single client which has the full file at any given time, provided that all sections of the file exist on at least one of the peers signed onto the tracker.