Naming and Service Discoveryin Peer-to-Peer NetworksECE1770 Expert TopicEli FidlerVinod MuthusamyFebruary 13, 2003OutlineTraditional Distributed Naming SystemsDistributed Naming ParadigmsP2P NamingExisting SystemsEmerging SystemsTraditional Naming SystemsTCP/IP Host NamingStatichosts filesNo central authorityHierarchicalDomain Name System (RFC1034/5)Authority for domains is delegated, but top level is centralizedCaching is vital for acceptable performanceDistributed Naming ParadigmsHost IDs (CORBA Naming Service)Each host is given a globally unique IDHosts are organized into hierarchical namespacesService IDs (CORBA Trader Service, Jini)Services are registered with broker, discovered using lookupDistributes Object IDs (file sharing networks)Each object has a unique ID, but may not exist in any single placeNode Discovery TechniquesStatic/NeighboursEach host has a static list of known nodes/neighboursCentralized RepositoryEach host knows the address of a repositoryLocal BroadcastA host searches for nodes using broadcast“Buddy List”A host connects to favourite/previously seen hostsP2P NamingStaticCentralizedNeighbour Discovery“Smart” DiscoveryEmerging Naming SystemsStatic P2P NamingEach host knows a static, explicit configuration of other nodesThe P2P network is staticc.f. hosts filesCentralized P2P NamingThere is a single host responsible for each service (or one host for all services)Nodes connect to P2P network, then contact host for desired serviceex. Napster, JiniJiniHierarchy of centralized lookup servicesAdvertisement = { interface name, attributes }Lookup = { interface name, [attributes] }Object moves from Provider to Lookup Service to ClientMust renew leasesPeer lookupLookup ServiceClientService ProviderService ObjectService Attributes1. Discover2. Join3. LookupNeighbour Discovery P2P NamingOnce connected to P2P network, hosts use P2P neighbours to connect to servicesSearches/commands propagate in wavesex. Gnutella/Limewire“Smart” Discovery P2P NamingOnce connected to P2P network, hosts use P2P neighbours to connect to servicesSearches/commands propagate along “best” path of neighbour-neighbour linksex. FreenetFreenetThe requests get routed to the appropriate host by incremental discoveryEmerging Naming SystemsTechnologiesJXTAIntentional Naming System (INS)Active Names AttributesNaming expressivenessArchitectureJXTASuper peers: distributed search hubsAdvertisement = { query space, predicates, address }Query = { query space, predicates }Groups of hubsEach group is responsible for some query space(s)Each group has a member from every other groupEach hub has a summary of adverts in every other hub in its groupSearch groupSearch hubINS – NamingName specifier ={ A hierarchy of attribute-value pairs }Name record = { Name specifier, metric, address }INS – ArchitectureINRs form spanning treeLate binding handles service/node mobilityName can refer to groupsScalability, load balancingActive NamesHierarchical namespace delegationActive Name = { name to resolve, namespace program }Namespace program = { Active Name }Service composition using after methodsLocation independent execution of namespace programSummary Decentralized administrationWell addressed Network failures, robustnessAddressed by periodic advertisementsAutomatic resolver spawning in INS LookupTypically need (distributed) servers (INS, Freenet, etc.)Flooding (Gnutella) is inefficientSummary (Cont'd) Query expressivenessPrimarily still hierarchical (INS, AN, Jini, etc.) Node/service mobilityAddressed by periodic advertisementsLate binding in INS ScalabilityMany are not scalable to Internet (INS, JXTA, Jini)Rely on lookup service hierarchy for