Peer-to-Peer Management of Large-Scale Memory Sources (midterm)

LOGO

XPeerXPeer

Peer-to-Peer Management of Large-Peer-to-Peer Management of Large-Scale Memory SourcesScale Memory Sources

Vincent Andrieu – Dr. Mark Roantree

Why is XPeer needed ?Why is XPeer needed ?

CDVP experiments so as to study the links between movie content and viewer reactions :

29 movies, 33 experiments, 44 persons and8 sensors per experiment

Sensors data stored in XML files that CDVP needs to query

250 XML files ; more than 670 MB Centralized XML database is not suitable Spreading the XML files over several databases and

querying them through a Peer-to-Peer (P2P) network.

Application design diagramApplication design diagram

P2P Network

QueryPeers

DataPeers(running on eXistservers (native XML databases))

ImplementationImplementation

XML Documents (DataPeers) gathered into clusters with keywords

ClusterPeer

Metadata stored in an Oracle database, containing info about all the peers

RepositoryPeer

4 peer types :• DataPeer• QueryPeer• ClusterPeer• RepositoryPeer

Implementation DiagramImplementation Diagram

LaunchingLaunching

Peers have a lot of attributes : ID Name Description Keywords Oracle/eXist URLs and username/password to connect Clusters they belong to ...

All this metadata is stored in the Oracle metabase.

When a peer is launched, the first thing it does is finding a repository peer (because it is connected to the metabase) so as to get his context, so as to know everything about itself.

• If a repository peer can’t find any other repository peer, then it connects to the metabase

• But if a non-repository peer can’t find a repository, it dies

JXTA platformJXTA platform

The JXTA platform is an open source peer-to-peer protocol specification.

Logical layer over the network layer

The physical network configuration of the peer is transparent. All of them are in the same logical P2P network, even if they actually are in different physical networks.

Connection and disconnections are managed.

JXTA platformJXTA platform As a JVM is very heavy and several peers may run on the same

machine, so the application was designed so as to allow several peers (even different types of peer) to run in the same process, over 1 JXTA Peer.

ClusterPeerDataPeerDataPeer

JXTA Platform

RepositoryPeerQueryPeer

JXTA Platform

DataPeerQueryPeerClusterPeer

JXTA Platform

P2P NetworkP2P Network

Peer designPeer design

What’s next ?What’s next ?

Caching systems (e.g. between ClusterPeer and DataPeer, between QueryPeer and

ClusterPeer)

Logical expressions or regexps for keywords (e.g. «cdvplex AND Shrek»)

Dynamically set the keywords for clusters