BearNotes

As of now, the specs say that if a client has a partial of a file and they intercept a request for that file that they can fulfill, then they take over the message and start sending their version of the file.

We think this could cause a number of problems since there is no way to positively identify a file as the file the user wanted. In the event that a user requests an entire file, there is no filesize involved, so anyone along the path who has a file with the same filename can respond, even if its not the file that was supposed to be sent.

While it makes sense for the first peer to serve the file, there is not enough information to make certain that indeed is the file.

This is also a case of file names existing as unique identifiers. But also, the protocol specifies that the FILE message must include the file size, offset from the beginning, segment length, and the filename.

NOTE: It will be up to the client to maintain some sort of data that indicates the portion of the file that they have cached.

On a related note, answers to search queries should only include complete files. It can then be up to the searching client what segment of the file that they need, and can request accordingly with a GET message.

Jason and Kirk