4  Internships

• Secure peer-to-peer collaborative backup systems
• Peer-to-peer systems monitoring

4.1  Secure peer-to-peer collaborative backup systems

Advisors: Anne-Marie Kermarrec (INRIA Rennes) et Fabrice Le Fessant (INRIA-futurs)

Keywords: peer-to-peer, backup

As the scale of distributed systems and applications keep growing, the peer to peer communication paradigm somehow has imposed itself as the key to scalability. In a peer to peer system, each peer may act both as a client and server so that the number of servers increases linearly with the size of the system thus ensure the scalability of the system as a whole. Resources are also aggregated between all the peers thus increasing the storage capacity. In a peer to peer system, the control is fully decentralized, there is no intrinsic bottleneck at the server as it is often the case in client/server system. Finally the last characteristic of a peer to peer system is that each peer has a local knowledge of the system and takes individual decision and yet the system as whole converges towards global properties

Many collaborative applications may benefit from such a generic scalable support. In this project we are interested in distributed collaborative backup systems where data from a given peer are backed up on a set of other peers.

The objective of this Ph.D is to design, evaluate and potentially deploy a collaborative peer to peer backup systems in which peers back up their data on a set of other peers. We are aiming at a reliable yet efficient storage system where confidentiality is ensured. Although there has been a lot of work in this area and that storage and archival system have been considered early on there are still many issues that have to tackled to be deployed. Among these issues, we have:

• Confidentiality
• Choice of peers on which the data have to be backed up depending on the data characteristics as well as target peers characteristic
• Performance
• Reliability
• Sharing capabilities
• Incentives

4.2   Peer-to-peer systems monitoring

Advisors: Fabrice Le Fessant (INRIA-futurs)

Keywords: peer-to-peer, file-sharing, monitoring, analyses, traces, Edonkey, Fasttrack

The goal of this work is to provide a better understanding of the behavior of users in peer-to-peer networks, by monitoring their behavior in existing networks, especially file-sharing networks which are the most used ones.

The implementation part of this study would be done in Objective-Caml, on the MLdonkey platform (a GPL multi-network client), either to extend it to monitor networks, or to analyse existing traces we have already collected.

We aim at collecting the following kind of information:

• Sharing patterns: what users share and how long (for example on Edonkey)
• Discovery queries and replies: how users query for files and which replies match these queries (for example on Kazaa).
• Downloading patterns: what users download, how long it take, with who the interact on the networks (for example on Bittorrent).
• Availability patterns: when users are connected to the network, and when their ressources are available (for example on Kademlia)

  We already have collected some traces, which could be analysed during an internship:

• traces of the Edonkey network containing, for each connected peer, the list of files it shares.
• traces of the Kazaa network containing the interactions between a peer and the supernode it is connected to (queries, sharing, etc)