Seminar presented 7 March 2005

1. CSM13CSM13
Software AgentsSoftware Agents
Seminar 5: Agents and P2P NetworksSeminar 5: Agents and P2P Networks
j.salter@surrey.ac.uk
http://www.computing.surrey.ac.uk/personal/st/N.Antonopoulos

2. Acknowledgements
Some of the slides in this presentation are taken or adapted from:
Sebastien Baehni. “Introduction to P2P and the JXTA framework”
http://www.ece.rutgers.edu/~parashar/Classes/03-04/ece451-566/slides/p2p.pdf
Frans Kaashoek et al. “Peer-to-Peer computing research: a fad?”
http://project-iris.net/talks/dht-toronto-03.ppt
Vinod Muthusamy. “An Introduction to Peer-to-Peer Networks”
http://www.eecg.toronto.edu/~jacobsen/mie456/slides/p2p-mie.pdf

3. Network Architectures
Client/Server Pure Peer-to-Peer
• Clients send requests to/via a
central server
• Small #messages
• Single point of failure
• Problem: Extreme scalability
• No central server
• Clients connected to
one or more peers
• Resilient
• Large #messages

4. Hybrid Peer-to-Peer Networks
• Some nodes act as group managers/routers
• More robust than client/server
• Potentially lower #messages than pure P2P

5. Characteristics of a P2P Network
• Nodes can act as clients and servers
• No centralised server/authority
– In Pure P2P Networks
• Network is highly dynamic
– Nodes join and leave regularly
• Large-scale
– Potentially millions of nodes
• Nodes are autonomous
– But co-operate to share/retrieve resources

6. Benefits of P2P Networks
• Efficient use of resources
• Scalability
• Reliability
• Ease of administration
but
• Increased message costs
• More complicated than client/server interaction

7. The Agent Relationship
• How does this relate to agents?
– P2P network platform for mobile agents
– Access agents over P2P network
– Agents administer P2P network
• P2P Networks are useless if agents, users and
applications cannot find resources they require
– Queries can be represented as agents

8. Research Question
How can we efficiently and accurately discover resources and
services in a P2P network?
Solution 1: Introduce some centralisation
Solution 2: Introduce some structure
“centralisation” and “structure” define two dimensions for classifying P2P networks

9. Napster
• Lists of files are uploaded
to Napster server
• Queries contain various
keywords of required file
• Server returns IP address
of user machines having
the file
• Client chooses user
machine and sets up
connection to it
• File transfer is direct

10. Napster
Centralised model
Napster server ensures correct results
Only used for finding the location of the files
Scalability bottleneck
Single point of failure
Denial of Service attacks possible
Lawsuits

11. Gnutella
• Decentralised search – no
central server
• Queries are sent to the
neighbour nodes
• Neighbours ask their own
neighbours, etc.
• Time To Live (TTL) field to
prevent loops
• File transfer is direct

12. Gnutella
Decentralised model
No single point of failure
Less susceptible to denial of service
SCALABILITY (flooding)
Cannot ensure correct results: “query
horizon”

13. KaZaA
• Hybrid of Napster and Gnutella
• Super-peers act as local search hubs
– Each super-peer is like a constrained Napster server
– Automatically chosen based on capacity and availability
• Lists of files are uploaded to a super-peer
• Super-peers periodically exchange file lists
• Queries are sent to super-peers

14. Freenet
• Ensures anonymity
• Decentralised search
• Queries are sent to the
neighbour nodes
• Neighbours ask their
own neighbours, etc.
• The query process is
sequential
• Learning ability
• Files propagate back
along query route

15. • Second generation P2P (overlay) networks
• Self-organizing
• Load balanced
• Fault-tolerant
• Guarantees on numbers of hops to answer a query
• Based on a distributed hash table interface
Structured P2P

16. • Distributed version of a hash table data structure
• Stores (key, value) pairs
• Goal: Efficiently insert/lookup/delete (key, value) pairs
• Each peer stores a subset of (key, value) pairs in the system
• Core operation: Find node responsible for a key
• Map key to node
• Efficiently route insert/lookup/delete request to this node
Distributed Hash Tables (DHT)

17. • Node id: m-bit identifier (similar to an IP address)
• Key: sequence of bytes
• Value: sequence of bytes
put(key, value)
 Store (key,value) at the node responsible for the key
value = get(key)
 Retrieve value associated with key (from the appropriate node)
DHT Generic Interface

18. • File sharing
• Databases
• Service discovery
• Chat service
• Publish/subscribe networks
DHT Applications

19. • Keys mapped evenly to all nodes in the network
• Each node maintains information about only a few
other nodes
• Efficient routing of messages to nodes
• Node insertion/deletion only affects a few nodes
DHT Desirable Properties

20. Node id: m-bit identifier (similar to an IP address)
Key: m-bit identifier (hash of a sequence of bytes)
Value: sequence of bytes
API
• insert(key, value)
• lookup(key)
• update(key, newval)
• join(n)
• leave()
Chord API

21. Chord Operation
• Nodes form a circle
based on node
identifiers
• Each node is
responsible in storing a
portion of the keys
• Hash function ensures
even distribution of
keys and nodes in the
circle

22. Chord Operation

23. Chord Operation
• Lookup the furthest
node that precedes the
key
• Query reaches target
node in O(logN) hops

24. Chord Properties
In a system with N nodes and K keys:
Each node manages at most K/N keys
Bounded routing information stored in every
node
Lookups resolved with O(logN) hops
Poor network locality

25. Network Locality
Nodes close on ring can be far in the network

26. Issues in P2P/Agents
• P2P applications cannot (usually)
communicate with each other
• Agent collaboration over P2P networks
• Enhancements to existing P2P protocols using
agents
• Enhance agent co-ordination schemes using
P2P co-ordination mechanisms
• Self organisation of P2P networks

27. Summary
• P2P can be used for sharing resources
• Removes need for central management
• Large scale, dynamic networks
• Efficient information organisation and retrieval
• P2P can enhance agent systems
• Agents can enhance P2P
• Focus gradually moves towards the coordinated
use of versatile, distributed computing resources