Lecture: Software Agents and P2P
Download as PPT, PDF0 likes595 views
This document discusses peer-to-peer (P2P) networks and their relationship to software agents. It describes different types of P2P network architectures and challenges with resource location. It proposes using a distributed hash table with a "ring of Chord rings" structure to index resources across keyword rings. Agents could route queries and help organize this P2P network to improve scalability, fault tolerance, and the ability to find resources. The document also notes opportunities to use agents and P2P together to enhance applications and coordination.
Lecture: Software Agents and P2P
- 1. CSM13: Software AgentsCSM13: Software Agents Agents and P2PAgents and P2P James Salter [email protected] 19 March 2004
- 2. OutlineOutline What is Peer-to-Peer? Relationship to Agents Locating resources in P2P Networks Problems with Napster and Gnutella Solutions Other Issues with P2P & Agents
- 3. What is “P2P Computing”?What is “P2P Computing”? Everybody seems to have a different idea File sharing Distributed computing Inter-connected communities of users Hideouts for virus writers and associated criminals Opposite of client/server network All nodes can function as clients and servers These are all valid definitions
- 4. Network ArchitecturesNetwork Architectures Client/Server Pure Peer-to-Peer Clients send requests to/via a central server Small #messages Single point of failure No central server Clients connected to one or more peers Resilient Large #messages
- 5. Hybrid Peer-to-Peer NetworksHybrid Peer-to-Peer Networks Some nodes act as group managers/routers More robust than client/server Potentially lower #messages than pure P2P
- 6. Characteristics of a P2P NetworkCharacteristics 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
- 7. The Agent RelationshipThe 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. Searching for ResourcesSearching for Resources Napster Hybrid P2P Queries sent to central server Nodes upload list of files they hold Server returns IPs of nodes hosting matching files Direct connection between nodes for transferring files File Transfer
- 9. Searching for ResourcesSearching for Resources Gnutella Pure P2P Network Queries flooded from node to node Query has Time-To- Live (TTL) counter Decremented each time visits a node Query discarded when TTL=0 Prevents endless loops
- 10. The ProblemThe Problem Scalability Napster: central server is a bottleneck Gnutella: large volumes of query traffic Number of Messages Gnutella: query must be sent to every node Fault Tolerance Napster: single point of failure Ability to Find Resources Gnutella: query horizon (TTL limit)
- 11. SolutionSolution Combine benefits of Gnutella and Napster Scalability and Number of Messages some form of structure to avoid visiting every node but more than one point of contact Fault Tolerance structure to avoid central server Ability to Find Resources provide alternative structure to remove TTL Introduce Structure
- 12. Hash TablesHash Tables Hash functions map data keys to buckets Keys are stored in buckets in index table Example: Distributed Hash Tables Each node hosts part of the index (one or more buckets) 0 1 2 3 4 5 f(x) = x mod 6 f(15) = 15 mod 6 = 3 12, 24, 30 19, 61 20, 26, 56 3, 9 10, 16, 28, 34 23, 35, 65 f(36) = 36 mod 6 = 0
- 13. Chord: ArchitectureChord: Architecture Nodes hosting part of the index table organised in a ring 1 41 7 22 28 34 48 52 70 73 Each node is given an identifier Hash function determines node host for a key(word) Nodes provide pointers to machines hosting resources
- 14. Chord: QueryingChord: Querying Queries routed around the ring following finger table information Finger table stores more information about nodes closer to it than those further away Routing ~ O(log N) Search space halved every hop 1 41 7 22 28 34 48 52 70 73
- 15. QueriesQueries Queries may be a list of keywords: “ford AND fiesta” We may need more complex queries, specifying keyword-value pairs: “make=ford AND model=fiesta AND mileage=20000” Boolean Operators: AND, OR, NOT keyword keyword value
- 16. A Ring of Chord RingsA Ring of Chord Rings Create a ring for every keyword Super Ring stores pointers to Keyword Rings Super Ring Keyword Ring Keyword Ring Keyword Ring Keyword Ring Keyword Ring
- 17. A Ring of Chord RingsA Ring of Chord Rings Support for name-value queries Each physical machine can take part in multiple rings Each node in super ring runs an agent responsible for keyword ring organisation and maintenance Agents on each node handle incoming queries Queries can be spawned as agents Multiple keyword queries (CPU=1.8 AND RAM=128) More complex queries?
- 18. Further ImprovementsFurther Improvements Record address of keyword rings Can route queries for previous keywords directly to keyword ring Cache previous search results Direct mapping between providers and consumers Cache becomes Preference List Ordered list of resource providers most likely to provide the required resource
- 19. Other Issues in P2P/AgentsOther 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 Micro-payments for resources/brokers
- 20. SummarySummary P2P can be used for sharing resources Removes need for central management Large scale, dynamic networks P2P can enhance agent systems Agents can enhance P2P Resource Discovery research area Agents and P2P can be closely related
Editor's Notes
- #4: File Sharing – virtually unlimited music, movies, software and “other” material all available for free 24/7. E.g. Napster (a few years ago), Gnutella, Kazaa, Limewire… Distributed Computing – harnessing unused processing power of millions of computers. E.g. the SETI@home project searching for aliens. 4.9million users, donated 1.8million years worth of CPU time. Inter-connected communities of users. E.g. Instant Messaging. MSN Messenger has 100million users worldwide. Hideouts for virus writers – Slapper one of the first viruses to build a peer to peer network of machines that creators can utilise – e.g. to launch a Denial of Service Attack. Opposite of a client/server network – machines can act as servers and clients. All valid definitions – but let’s look at the last one in more detail
- #5: In traditional client/server environment all requests are handled by the server. No direct client to client communication. There are only a small number of messages required to send data to the server. Each machine knows where the server is on the network. Having a single server leads to a single point of failure. If the server goes down, nobody can access any resources. Opposite of this is a pure peer-to-peer environment. Each machine can be both a client and server, so we remove the need for a single central server and therefore the single point of failure. Clients no longer know exactly where the machine hosting the resource they need is. Increase in #messages required to find the resource.
- #6: Not all P2P networks “pure”. In fact, many are hybrids somewhere between client/server and pure P2P. More organisation – group managers hold more information about the network so resources can be found in less messages.
- #7: Notice many of the characteristics are similar to agents
- #8: Agents can communicate over P2P - autonomous agents more suited to autonomous environments than client/server.
- #9: Napster had a central server where all search queries were sent. Server held list of all machines that hosted the requested file. Direct connection between peers only set up to actually download the file. The single centralised server ultimately led to Napster’s downfall. Lawyers for the Recording Industry Association of America (RIAA) were able to show Napster knew their users were sharing copyrighted material (and could stop it) because everything passed through their servers. When the central server was shut down, the Napster network was broken.
- #10: Huge amounts of traffic can be generated if many users are submitting queries at the same time What happens if a machine hosting the file you are searching for lies outside the query “horizon”? – It will not be found.
- #13: Developed in 1960s Like pigeon holes in the common room.
- #14: Nodes organised in ring according to the identifier they have been given. Could be generated by hashing their IP address. Nodes host portion of hash table around their identifier. The node hosting the index is not necessarily the node actually providing the resource – separation of index from resource (like Napster, unlike Gnutella)
- #16: What do queries look like?
- #17: Improve over existing Chord No need to route through a single large ring containing every node – can route using less messages.
- #19: Record address of keyword rings as queries are conducted: Reduces #messages necessary to route query through ring. Improves fault-tolerance – these queries no longer rely on ANY node in Super Ring. Cache previous results: no reliance on indexing structure Preference List: further reduce #messages required to find required resource. Whether resource exists on located node may not be only consideration: Is node regularly too busy to provide resource? Is node willing to provide you with the resource? Do you have compatible security policies?
- #21: P2P can be used for file sharing, distributed computing, creating communities – more generally sharing resources. No need for centralised indexes or servers – this also means the network must be self organising. Not talking about company networks – looking at worldwide scale networks (think Internet) P2P can enhance agent systems: Agents can communicate over P2P. Agents can be nodes at edges of P2P network. Agents can enhance P2P: Help in management of network (P2P networks must be self-organising – no central server). Agents can be used to discover resources. Resource Discovery in P2P networks is an ongoing research area. Agents can be useful in this. Current research activities looking at many uses of P2P and agents.