Research On Key Technologies Of P2P-based Semantic Publish/Subscribe Systems

With the rapid development of Internet, mobile devices and broadband networks,network computing is getting more and more complex. The main characteristics of thenew generation of network computing are large-scale, decentralized control, dynamic,and autonomous. Since publish/subscribe systems have the advantages of loosecoupling, anonymity, many-to-many communication and scalability, they have becomeimportant infrastructure middleware platforms support for the new generation ofnetwork computing.Publish/subscribe systems have gained fairly mature research results oncontent-based data models, matching algorithms and routing algorithms, but they cannot yet meet challenges posed by the new generation of network computing, mainlydue to the lack of support for semantic events and semantic routing. In order to providepowerful middleware support for the new generation of network computing, this thesismakes the research on key technologies including semantic data model, semanticmatching algorithm, and semantic routing algorithm based on structural P2P networks,and develops a prototype of semantic publish/subscribe system called JTangPS (JTangSemantic Publish / Subscribe System). The work has been supported by the NationalHigh-Tech. R&D Program, China.In the first part of thesis, we discuss the research background and significance ofpublish/subscribe system. After explaining the basic publish/subscribe system modeland comparing it with relevant communication models, we introduce key technologyresearch situation on data model, matching algorithm and routing algorithm, summingup the advantages and disadvantages of various methods. Then we analyze therequirements of publish/subscribe systems proposed by the new generation of networkcomputing and introduce the research hotspots of publish/subscribe systems.In the second part, we propose a component-based layered architecture forsemantic publish/subscribe system. After analyzing some typical publish/subscribesystem architectures, we introduce various design frameworks for publish/subscribesystems and discuss basic implementing technologies for JTangPS under the givenframework. At last, we present the architecture of JTangPS on two different levels ofstructures, namly layered architecture and concrete implementing structure. JTangPShas a loosely coupled architecture in which system components can be replaced byones with the same functions but different implementations, thereby quickly meetingthe demands of various application scenarios. In the third part, we propose a new semantic data model based on OWL (WebOntology Language) and RDF (Resource Description Framework). The semantic datamodel includes conceptual model, event model and subscription model, which is thebase of understanding semantic information for publish/subscribe systems. Theconceptual model uses OWL.language; the event model uses RDF graph; and thesubscription model uses RDF graph pattern. The concept model can directly employexisting OWL ontologies on networks, and ordinary RDF graphs can be treated asevents. The semantic data model of JTangPS solves the problem on data withheterogeneous syntactic structures but homogeneous semantics, so that events andsubscriptions can be understood and dealt by machines without ambiguously. Inaddition, we propose a new subscription language RESL (RDF Event SubscribeLanguage) which is similar to existing RDF query languages.In the fourth part, we propose a fast RDF graph based matching algorithm.Efficient algorithm is a key technology to reconcile rich expressiveness with systemscalability. The basic idea behind the semantic matching algorithm of JTangPS is:devising an event or a subscription into a series of arcs of which each is the basicmatching unit; creating indexes for subscriptions and events to make full use of theoverlap between subscriptions and narrow the matching scope; adding the equivalentsand ancestors of attribute to index structures and being combined with the type checkof variable bindings of subscriptions to achieve semantic matching; doing natural joinoperations on variable binding tables to eliminate unnecessary constraint checks,thereby improving event matching efficiency. Experimental results demonstrate thatthe algorithm is superior to G-ToPSS in performance and much better than SMA(Simple Matching Algorithm) which simply matches events with each subscription.In the fifth part, we propose a rendezvous-based semantic routing algorithm onstructured peer-to-peer networks. DHT-based P2P networks are so self-organizing,fault-tolerant and scalable that not only it adapts to the dynamic changes of networks,but also guarantees the resources discovery accuracy, hence being very suitable as thesubstrate of publish/subscribe system. The basic idea behind the semantic routingalgorithm of JTangPS is to map events and subscriptions to rendezvous according tothe combination of domain identifier, the number of attributes and attribute name.Mapping attribute name not only solves the conflict between precision in data mappingand complexity of data model on DHT-based P2P networks, but also avoids theconstraints on subscription language while mapping subscriptions to rendezvous;attribute number controls the number of rendezvous and limits the destinations towhich events are published, thereby reducing the traffic of event publication and avoiding unnecessary event matching computation; it makes full advantage offault-tolerance of P2P and reduces the routing flow for event delivery to employ theembedded routing mechanism of P2P and the aggregation optimization approach.Experimental results show that for large-scale publish/subscribe, the semantic routingalgorithm of JTangPS is superior to the reverse-path based forwarding routingalgorithm and achieves a good balance among routing efficiency, network resourceconsumption, subscription maintenance efficiency and system scalability.In the sixth part, we explore the implementation of JTangPS prototype system andintroduce the practical application of JTangPS by the example of selectivelydisseminating RSS feeds, which demonstrates the system architecture, the semanticdata model, the semantic matching algorithm and the semantic routing algorithmdiscussed in previous parts.