1. OntoZilla: An Ontology-based, Semi-structured, and Evolutionary P2P Network for Information Systems and Services 指導教授：李官陵 學 生：陳建博 69621501 蔡英傑 69621515

2. 2 Outline 1. Introduction 2. OntoZilla 2.2 Semantic Links 3. Construction & Maintenance 4. Conclusions 3.1 Join 3.2 Routing 2.1 System Overview 3.4 Merging 3.5 Leave 3.3 Stabilization

3. 3 1. Introduction – Why use Ontology? 1)Current keyword search methods can retrieve huge irrelevant data that contain certain terms with different meanings. However, they may miss relevant data that contain different terms with the same meaning. 2)SO, current P2P lack of semantic capability hinders the advancement of P2P systems and makes them only adopted in several fields, such as file-sharing. 3) Ontologies which express knowledge and the relationships in the knowledge with clear semantics, thus improve the process of information searching.

4. 4 1. Introduction – P2P combined with Ontology 1)The proposed approach was designed to utilize the synergy between ontologies and P2P systems, and could improve the process of information searching. 2)In a P2P system combined with ontologies, users can annotate their resources with ontologies, can make resource machine-interpretable and easily share the resources with each other via the overlay network. 3)To make heterogeneous information would be automatic processing,and smooth the way to greater integration and interoperability.

5. 5 1. Introduction – OntoZilla 1)Peers in OntoZilla are organized according to their relationship which modeled with ontologies are based on their interests. 2)In a P2P system, a peer may have interests in some kinds of information, or concentrate on providing particular expertise or services. This characteristic can be viewed as the Special Interests of a peer. 3)Using semantic links to reflect the relationships among peers such as hierarchical or cooperative relationship, queries can be routed to the peers with suitable information or services in a systematic way.

6. 6 2. OntoZilla – 2.1 System Overview 1)Special Interest Groups (SIGs).  In a P2P system, a peer may concentrate on providing particular expertise or services. This characteristic can be viewed as the Special Interests of a peer.  Peers' special interests can be categorized into groups, referred to as Special Interest Groups (SIGs). Peers with identical or similar special interests are grouped into the same SIG.  The categorization can refer to the first-level classes of some classification systems, for example, Dewey Decimal Classification and Yahoo! Directory.

7. 7 2. OntoZilla – 2.1 System Overview 2)Clusters and Semantic Links  Within each SIG, peers employ some classification systems to hierarchically classify their special interests into concepts, or classes.  Each class supporting the same concept within each SIG can be created as a cluster that further groups peers supporting the same concept.  Peer relationships can be modeled with Ontologies. These relationships could be designed as several kinds of links with semantics, or semantic links.

8. 8 2. OntoZilla – 2.1 System Overview  parent links and child links - Peers that have an inheritance relationship can connect with each other  twin links – peers in the same cluster can connect with each other.  SIG links - peers in different SIGs can connect with each other.  partner links - peers that have a cooperative relationship can connect with each other  As a result, peers supporting the same concept form a cluster, and all clusters in the same SIG form a hierarchical tree.

9. 9 2. OntoZilla – 2.1 System Overview 2)Classification Systems  In a classification hierarchy, the link between a parent and child node denotes an “is subclass of” (is-a) relationship.  The classes that have direct is-a relationships with a lower class are called its parent classes, whereas the classes that have direct is-a relationships with a higher class are called its child classes.  The classes that have indirect is-a relationships with a lower class are called its ancestor classes, whereas the classes that have indirect is-a relationships with a higher class are called its descendant classes.

10. 10 2. OntoZilla – 2.1 System Overview  Parent and ancestor classes are all called super-classes, whereas child and descendant classes are all called subclasses.  For an illustration purpose, in this paper we use “Music”, “Business”, and “Computers” as the sample SIGs, and the ACM Computing Classification System (CCS) as the classification system for the “Computers” SIG.  Hence, in the “Computers” SIG, the name of the root class is “Computers (0)”, and its child classes are “Hardware (1)”, “Computer systems organization (1)”, “Software (1)”, “Computing methodologies (1)”.

11. 11 2. OntoZilla – 2.1 System Overview 3)Ontology-based Descriptions  Each SIG has a SIG description, which contains the SIG name and some annotations of the SIG. SIG descriptions are known by all peers in the network.  Each class has a class description containing the class name, the SIG name, the classification hierarchy, and some annotations of the class.  A peer description contains the peer’s SIG name, SIG ID, class name, cluster ID, IP address, along with some additional information that can be used to judge the adequacy of the peer.

12. 12 2. OntoZilla – 2.1 System Overview  SIG and class descriptions are represented with ontology languages so that relationships between SIGs and/or classes can be inferred with a predefined ontology.  By using some semantic matching techniques, we can also measure the similarity between SIGs or between classes if the similarity measure exceeds a specific matching threshold.  This evaluation is important for a new peer to determine which SIG and class to join, as well as for peers to build their relationships.

13. 13 2. OntoZilla – 2.2 Semantic Links 1)Based on peer descriptions, A peer establishes semantic links to other peers by simply adding their peer descriptions into its routing table. Semantic links can be categorized into the following three types:  SIG links - which are used to connect the peers of different SIGs.  Family links - which are used to connect the peers of the same SIG.  Partner links - which are used to establish a cooperative relationship between peers.

14. 14 2. OntoZilla – 2.2 Semantic Links Family links can be further divided into the four types:.  Twin links: Peers belonging to the same class are grouped into a cluster.  Parent links and Child links: A peer connects to peers of its parent class with parent links, whereas it connects to peers of its child classes with child links.  Ancestor links and Descendant links: Not all clusters may exist, peer may establish direct connections with the peers of its closest ancestor class or closest descendant classes.  Cousin links: Peers in the root clusters of disjoint subtrees establish cousin links with each other.

15. 15 2. OntoZilla – Sketch of Ontozilla

16. 16 2. OntoZilla – 2.3 An ontology-based, semi-structured, and Evolution network 1)Ontology-based  Each peer joins the clusters or SIGs that are semantically close to its supporting concepts or special interest.  Peers in the network use semantic links to reflect their relationships. 2)Semi-structured  Peers supporting the same concepts form a cluster, and all clusters in the same SIG form a hierarchical.  A peer may join more than one SIG. Moreover, the communication channels between different SIGs are also established.

17. 17 1)Evolution network  Since peers may join and leave the network at will, peers have to update their relationships as the network evolves.  family link priorities- The priority of parent links is higher than that of ancestor links. The priority of child links is higher than that of descendant links. The closer the ancestor class’s level, the higher the ancestor link’s priority. The closer the descendant class’s level, the higher the descendant link’s priority 2. OntoZilla – 2.3 An ontology-based, semi-structured, and Evolution network

18. 18 3. Construction and Maintenance – 3.1 Join 3.1.1 SIG Join Process – SIG inquiry, SIG answer  First, peer x randomly selects a peer, say y, as the contact peer and connects to the network.  Next, x broadcasts SIG inquiry messages through y within the range MAX_HOP_SIG_INQUIRY.  Each SIG inquiry message contains x ’ s SIG name and IP info …  All the peers that receive the messages will reply to x with SIG answer messages that contains some IP info of peers in SIG x.  If x receives SIG answer messages of the same SIG, and joins this existing SIG with the help of z of same SIG, Otherwise x creates a new SIG that most matches its special interests.

19. 19 3. Construction and Maintenance – 3.1 Join 3.1.1 SIG Join Process – peer join  Then, peer x broadcasts peer join messages which carry its description with other peers within the range MAX_HOP_PEER_INQUIRY.  All the peers that receive the messages are informed of x’s participation in SIG x, and send their peer description to x. 1)The peers of different SIG x then add SIG links to x into their routing tables, establish the communication channels to SIG x. 2)The peers of SIG x can establish or replace their family links to x according to the family link priority.

20. 20 3. Construction and Maintenance – 3.1 Join 3.1.1 SIG Join Process – example  First, a existed peer a, whose SIG is “ Music ”, had created a “ Music ” SIG in the network.  Next, peer b, whose SIG is “ Business ”, joins the network by using peer a as its contact peer. Then, b broadcasts SIG inquiry messages through a within the range MAX_HOP_SIG_INQUIRY.  As b cannot find the “ Business ” SIG, it creates a new SIG. Following that, it broadcasts peer join messages through a within the range MAX_HOP_PEER_JOIN.  After a is informed of the creation of the “ Business ” SIG, it sends its peer description to b, and then adds a SIG link to b.

21. 21 3. Construction & Maintenance – 3.1 Join

22. 22 3. Construction & Maintenance – 3.1 Join 3.1.2 Class Join Process The class join process is divided into two stages: 1)Class Finding Stage Peer x has to find its own cluster. If x can ’ t find its cluster, it finds the cluster that is closest to its hierarchical position. x use z as “guide peer” to join its class cluster in the SIG. ‚x copies z’s routing table to itself. ƒThrough family links, x can visit clusters in turn. „Then, x joins a cluster in accordance with Algorithm 1.

23. 23 3. Construction & Maintenance – 3.1 Join

24. 24 3. Construction & Maintenance – 3.1 Join 2)Family Linking Stage In this stage, x has to establish family links with peers in the same SIG. According to Algorithm 1 in the class finding stage, peer x may encounter six cases.  In the family linking stage, the way to establish family links in Case 2 and 3 are similar. (Through its parent/ancestor cluster)  In the family linking stage, the way to establish family links in Case 4 and Case 5 are similar. (Through its child/descendant cluster)

25. 25 3. Construction & Maintenance – 3.1 Join 3)Example

26. 26 3. Construction & Maintenance – 3.1 Join Case 1 (x join through its own cluster)

27. 27 3. Construction & Maintenance – 3.1 Join Case 2 (x joins through its parent cluster)

28. 28 3. Construction & Maintenance – 3.1 Join Case 4 (x joins through its child cluster)

29. 29 3. Construction & Maintenance – 3.1 Join 3.1.3 Fault Tolerance 1)Rule1 Only when the number of links a peer uses to connect to a specific cluster is greater than or equal to MIN_LINK_TO_ CLUSTER, can the peer drop its links to another cluster with a lower family link priority. 2)Rule2 When the number of peers in a specific cluster is less than MIN_LINK_TO_ CLUSTER, all the peers that have links to the cluster have to establish links to another cluster with a lower family link priority.

30. 30 3. Construction & Maintenance – 3.2 Query routing 1)Intra-SIG Routing  If a peer receives a query that requests resources belonging to its own class, broadcasts throughout its cluster.  If a peer receives a query that requests resources belonging to its super-classes, forward to higher cluster.  If a peer receives a query that requests resources belonging to its subclasses, forward to lower cluster.  If the querying peer is not satisfied with the number of answers, it may issue a query to its subclasses. Still not satisfied adjust the class of the query to a higher level.

31. 31 3. Construction & Maintenance – 3.2 Query routing 2)Inter-SIG Routing  If a peer needs resources outside its own SIG, it looks up its routing table and forwards the query to a peer in the suitable SIG via SIG links.  If its routing table doesn’t contain any peer in the suitable SIG, it broadcast SIG inquiry messages within the range MAX_HOP_SIG_INQUIRY. (like SIG join process)  After the peer finds a peer in that SIG and sends the query to it then routed according to the intra-SIG Routing.

32. 32 3. Construction & Maintenance – 3.3 Stabilization 1)Peer Probe In OntoZilla, each peer regularly sends peer probe messages to its neighbor peers to probe their availability, and then drops those semantic links to unavailable peers. 2)Peer Gossip Peer gossip is the periodical exchange of peer descriptions within some range to allow peers to update their views of the network and to correct their family links if needed.

33. 33 3. Construction & Maintenance – 3.4 Merging 3.4.1 Cluster Merging  Due to concurrent join and propagation delay, disjoint clusters of the same class in a SIG may be created by different peers.  Cluster partition can be detected early in the family linking stage, or later on through peer gossip. When a peer discovers a cluster partition, it will notify some peers in the partitioned clusters to start the cluster merging process.  Principle: The one with a smaller cluster ID is the winner, while the other is the loser.  The loser cluster will be merged into the winner, and change its cluster ID to the new one.

34. 34 3. Construction & Maintenance – 3.4 Merging 3.4.2 SIG Merging 1)Two SIGs of different IDs may be created simultaneously due to same reasons like cluster partition. 2)Principle: the one with smaller ID will be the winner, while the other is the loser. 3)The loser SIG merge into the winner SIG, copy winner’s family links,and packs them into a SIG merging message. 4)All the peers in the loser SIG that receive the SIG merging message for merging their clusters into the winning SIG. Merge two partitioned cluster as like cluster merging process. ‚Insert a cluster into the winner SIG as like cluster join process.

35. 35 3. Construction & Maintenance – 3.5 Leave 1)Before a peer x leaves the network, it sends leave messages to all its neighbor peers. Each leave message contains some peer descriptions of x's twin neighbors. 2)Therefore, all the peers that receive the messages and have semantic links to x can replace their links to x with the links to its twin neighbors. 3)For peers that have semantic links to x but did not receive x's leave messages, they can still actively or passively recover the links to x as like peer gossip process.

36. 36 4. Conclusion 1)OntoZilla combines ontologies and P2P systems, with a vision of improving information searching and facilitating greater integration as well as interoperability. 2)It can be applied to many areas such as a platform for concept search, dynamic service discovery and so on. 3)The realization of our vision relies heavily on the maturity of ontology technologies. 4)We assume that peers in a SIG use only one classification system to classify special interests, information, and services.