File Sharing and Downloading

File Sharing and Downloading

Background

• Problem description
  • A persistent storage system stores files and replicas, using the peers' resources.
  • The underlying overlay network is used to locate and route files.
  • Existing P.S. systems provide only the rudimentary service of locating a single replica of a file.
  • We want to adapt the system to provide efficient file downloading from peers, with respect to the scenario.

Directions and questions

• Peer searching
  • P.S. systems route to the nearest replica of the requested file.
  • We need to search for as many replicas as we can find, in order to:
    • Increase the download speed;
    • Decrease the load on the peer holding the replica.
  • An idea: use a BitTorrent-style tracker which keeps a list of seeders for the file. Downside: centralized solution. Need to make it scalable.
• Using rateless codes
  • Rateless codes are a form of erasure codes which are capable of producing an unlimited number of blocks.
  • Using rateless codes to download files increases the chance of completing the download quickly.
  • We need see how to employ rateless codes when downloading from peers.
• Spontaneous multicast
  • There may be cases when many peers request the file at the completion phase, creating a load on the peers who have that file.
  • What we'd like is starting a "spontaneous multicast" ? peers who request the file will join a forming multicast tree, like in the multicast phase.
  • We need to see if "spontaneous multicast" can indeed be employed when "enough" peers request the file.
• Downloading related files
  • Some files are related and are expected to be found together on a peer's computer. For example: packages related through dependencies.
  • Using knowledge of related files can save a peer time and bandwidth when looking for those files on the network.
  • We need a method to represent a peer's knowledge of relation between files, and to use that knowledge to save time and bandwidth.

Summary of open issues

• Finding as many replicas as possible for a quick download.
• Employing rateless codes.
• Using spontaneous multicast ? when and how?
• Efficiently downloading related files.