FREE ELECTRONIC LIBRARY - Abstract, dissertation, book

Pages:     | 1 |   ...   | 18 | 19 || 21 | 22 |   ...   | 25 |

«Approximate Information Filtering in Structured Peer-to-Peer Networks Christian Zimmer Max-Planck Institute for Informatics Saarbrücken ...»

-- [ Page 20 ] --

7.1.3 Previous Research on P2P Digital Library Architectures P2P-DIET [IKT04a] and LibraRing where the first approaches that tried to support both IR and IF functionalities in a single unifying framework. P2P-DIET utilizes an expressive query language based on IR concepts and is implemented as an unstructured P2P network with routing techniques based on shortest paths and minimum weight spanning trees. An extension of P2P-DIET [CIKN04] considers a similar problem for distributing RDF metadata in an Edutella [NWQ+ 02] fashion. LibraRing [TIK05a] was the first approach to provide protocols for the support of both IR and IF functionality in DLs using DHTs. In LibraRing, super-peers are organized in a Chord DHT and both (continuous) queries and documents are indexed by hashing words contained in them. This hashing scheme depends heavily on the data model and query language adopted, and the protocols have to be modified when the data model changes [TIK05a]. The DHT is used to make sure that queries meet the matching documents (in the IR scenario) or that published documents meet the indexed continuous queries (in the IF scenario). In this way the retrieval effectiveness of a centralized system is achieved, while a number of routing optimizations (such as value proxying, content based-multicasting, etc.) are used to enhance scalability. [RPTW08] presents iClusterDL, a self-organizing overlay network that supports information retrieval and filtering functionality in a digital library environment. Contrary to approaches like LibraRing [TIK05a] that focus on exact retrieval and filtering functionality (e.g., by disseminating documents or continuous queries in the network), in MinervaDL publications are processed locally and query or subscribe to only selected information sources that are most likely to satisfy the user’s information demand. In this way, efficiency and scalability are enhanced by trading faster response times for some loss in recall, achieving approximate retrieval and filtering functionality. MinervaDL is the first approach to provide a comprehensive architecture and the related protocols to support approximate retrieval and filtering functionality in a digital library context. Contrary to the LibraRing approach, in MinervaDL the Chord DHT is used to disseminate and store metadata about the document providers rather than the documents themselves. Avoiding per-document indexing granularity allows to improve scalability by trading recall for lower message traffic. This approximate retrieval and filtering approach relaxes the assumption of potentially delivering notifications from every producer that holds in the works mentioned above and amplifies scalability. Additionally, it allows to easily support different data models and query languages, without modifications to the protocols, since matching is performed locally in each peer.

–  –  –

Figure 7.2: High-Level View of the MinervaDL Architecture.

7.2 The MinervaDL Architecture This section of the use case presents the high-level view of the MinervaDL architecture and presents the various system components. The system architecture of MinervaDL is composed of three different types of peers: super-peers, consumer peers (or consumers), and provider peers (or providers).

Figure 7.2 shows a high-level view using an underlying DHT-based directory (e.

g., Chord [SMK+ 01]). The following sections explain the three main components and explain their properties and abilities in detail while Section 7.3 presents the protocols regulating peer interactions.

7.2.1 Super-Peers Super-peers run the DHT protocol and form a distributed directory that maintains statistics (metadata) about providers’ local knowledge in compact form. In MinervaDL, the Chord DHT is used to partition the key space such that each directory peer (super-peer) is responsible for the statistics of a randomized subset of keys. Directory peers are super-peers, peers with more capabilities than consumer or provider peers (e.g., more cpu power and bandwidth capacities) that are responsible for serving information consumers and providers and act as their access point to the MinervaDL network. When the number of super-peers is small, each peer can easily locate others in a single hop by maintaining a full routing table. When the super-peer network grows in size, the DHT provides a scalable means of locating other super-peers in the network.

Super-peers can be deployed by large institutions like universities, research centers or content providers (e.g., CiteSeer, ACM, Springer, Elsevier) to provide access points for their users (students, faculty or employees) or digital libraries. As shown in Figure 7.2, more than one provider and/or consumer can be connected to a single super-peer that acts as their common access point.

- 125 Chapter 7 Digital Library Use Case

7.2.2 Consumer Peers Consumer peers (or consumers) are utilized by users (e.g., students, faculty or employees) to connect to the MinervaDL network, using a single super-peer as their access point. Utilizing a consumer peer allows users to pose one-time queries, receive relevant resources, subscribe to resource publications with continuous queries and receive notifications about published resources (e.g., documents) that match their interests. Consumer peers are responsible for selecting the best information sources to query (respectively monitor) with respect to a given one-time query (respectively continuous query). If consumer peers are not online to receive notifications about documents matching their submitted continuous queries, these notifications are stored by their access point and are delivered upon reconnection. Section

7.3 presents the protocols regulating the activities of consumer peers.

7.2.3 Provider Peer Provider peers (or providers) are implemented by information sources that want to expose their content to the MinervaDL network. Typical examples are digital libraries deployed by larger institutions, like research centers or content providers (e.g., CiteSeer, ACM, Springer, or Elsevier). Provider peers use a directory peer (super-peer) as their access point and utilize it to distribute statistics about their local resources to the network. Providers answer one-time queries and store continuous queries submitted by consumers to match them against new documents they publish. More than one provider peers may be used to expose the contents of large digital libraries, and also an integration layer can be used to unify different types of DLs.

7.3 The MinervaDL Protocols Having introduced in the previous section the main architecture of MinervaDL with three different types of peers, in this section, the protocols that regulate the interactions between all types of peers in the DL architecture are explained in detail. The protocols include the joining and leaving of consumers, providers, and super-peers, but also the publication of new documents, the submission of one-time or continuous queries, and the receipt of answers and notifications for submitted requests.

Before explaining the individual protocols, three different functions have to be defined.

These will be used to ensure the basic communication procedures of the presented protocols:

–  –  –

7.3.1 Provider & Consumer Join/Leave The first time, a provider peer P wants to connect to the existing MinervaDL network, it has to follow the join protocol. P has to find the IP address of a super-peer S using out-of-band means (e.g., via a secure Web site that contains IP addresses for the super-peers that are currently online in the network). Then, P sends to S a NewProv(key(P ), ip(P )) message, and S adds P in its local provider table (P T ), which is a hash table used for identifying the providers that use S as their access point. Here, key(P ) is used to index providers in P T, while each P T slot stores contact information about the provider including its status (connected or disconnected) and its stored notifications (see Section 7.3.8 for notification delivery).

Subsequently, super-peer S sends to provider P an appropriate acknowledgement message AckNewProv(id(S), ip(S)). Once P has joined, it can use the connect/disconnect protocol described next to connect to and disconnect from the network. A consumers C uses a similar protocol to join the MinervaDL network. In this case the appropriate messages NewCons and AckNewCons are utilized in combination with a consumer table CT managing contact information about consumers.

A provider peer P or a consumer peer C that want to leave the network has to send a LeaveProv(key(P ), ip(P ), id(S)) or LeaveCons(key(C), ip(C), id(S)) message to its access point S, respectively. The super-peer S deletes the peer from its provider or consumer table including all contact information.

7.3.2 Provider & Consumer Connect/Disconnect When a provider P wants to connect to the network, it sends to its access point S a ConnectProv(key(P ), ip(P ), id(S)) message. If key(P ) exists in P T of S, P is marked as connected. If key(P ) does not exist in P T, this means that S was not the access point of P the last time that P connected (Section 7.3.8 discusses this case). When a provider P wants to disconnect, it sends to its access point S a DisconnectProv(key(P ), ip(P ), id(S)) message and S marks P as disconnected in its P T.

Consumers connect/disconnect from the network in a similar way (applying messages ConnectCons and DisconnectCons), but S has also to make sure that a disconnecting consumer C will not miss notifications about resources of interest while not online. Thus, notifications for C are stored in the consumer table CT of S and wait to be delivered upon reconnection of C (see Section 7.3.8).

7.3.3 Super-Peer Join/Leave To join the MinervaDL network, a super-peer S must find the IP address of another super-peer S using out-of-band means. S creates a NewSPeer(id(S), ip(S)) message and sends it to S which performs a lookup operation by calling lookup(id(S)) to find Ssucc = successor(id(S)), similarly to the Chord joining procedure. S sends a AckNewSPeer(id(Ssucc ), ip(Ssucc )) message to S and S updates its successor to Ssucc. S also contacts Ssucc asking its predecessor and the data that should now be stored at S. Ssucc updates its predecessor to S, and answers back with the contact information of its previous predecessor, Spred, and all continuous queries and publications that were indexed under key k, with id(S) ≤ k id(Spred ). S makes Spred its predecessor and populates its index structures with the new data that arrived. After that S populates its finger table entries by repeatedly performing lookup operations on the desired keys.

- 127 Chapter 7 Digital Library Use Case

When a super-peer S wants to leave MinervaDL network, it constructs a DisconnectSPeer(id(S), ip(S), id(Spred ), ip(Spred ), data) message, where data are all the continuous queries, published resources and stored notifications of off-line peers that S was responsible for. Subsequently, S sends the message to its successor Ssucc and notifies Spred that its successor is now Ssucc. Clients that used S as their access point connect to the network through another super-peer S. Stored notifications can be retrieved through successor(id(S)).

–  –  –

Subsequently, C creates a GetResults(ip(C), key(C), q) message and forwards it, using the contact information associated with the statistics, to all provider peers selected previously. Once a provider peer P receives a GetResults message containing a query q, it matches q against its local document collection to retrieve the documents matching q.

The local results are ranked according to their relevance to the query to create a result list R. Subsequently, P creates a RetResults(ip(P ), R, q) message and sends it to C. In this way, C collects the local result lists of all selected providers and uses them to compute a final result list that is then presented to the user. To merge the retrieved result lists, standard IR scoring functions (e.g., CORI [CLC95], GlOSS [GGMT99], or CVV [YL97]) are used. In [FPC+ 99], various standard approaches are compared.

7.3.6 Subscribing with a Continuous Query

This section describes how to extend the protocols of Section 7.3.5 to provide information filtering functionality. To submit a continuous query cq containing keys k1, k2,..., kn, the one-time query submission protocol needs to be modified. The first three steps are identical while step four is modified as follows.

C uses the scoring function pred(P, cq) described in Section 7.4 to rank the providers with respect to cq and identify the top − k providers that may publish documents matching cq in the future. These are the peers that will store cq and C will receive notifications from these peers only. This query indexing scheme makes provider selection a critical component of the filtering functionality. Notice that, in a filtering setting, resource selection techniques like sel(P, cq) described in Section 7.4 and used for one-time querying, are not appropriate since MinervaDL is not interested in the current document collection of the providers but rather in their future publishing behavior.

Once providers that will store cq have been determined, consumer C creates an message IndexQuery(key(C), ip(C), id(S), ip(S), cq) and sends it to these providers using the IP addresses associated with the GetStats messages C received in the previous step. When a provider peer P receives an IndexQuery message, it stores cq in its local continuous query data structures to match it against future publications. P utilizes these data structures at publication time to find quickly all continuous queries that match a publication. This can be done using efficient algorithms, e.g., BestFitTrie [TKD04], or SQI [YGM99].

7.3.7 Publishing a new Document

Pages:     | 1 |   ...   | 18 | 19 || 21 | 22 |   ...   | 25 |

Similar works:

«MISSIONARY ATLAS PROJECT EUROPE Gibraltar Snapshot Section Country Name: Gibraltar but also known as as Jabal Tariq since AD 711 Country Founded in: In 1713, Gibraltar became a dependency of Great Britain.Population: 27,967 (July 2007 est.) Government Type: Parliamentary representative democratic dependency. Geography/location in the world: 39 11 N, 5 22 W A part of Europe, Gibraltar is located at the southernmost tip of the Iberian Peninsula. It overlooks the Strait of Gibraltar. Number of...»

«Ex-Post-Bewertung des Agrarinvestitionsförderungsprogramms (AFP) im Förderzeitraum 2000 bis 2006 Länderübergreifender Bericht Verfasser: Bernhard Forstner (Einzelbetriebliche Wirkungen) Angela Bergschmidt (Umwelt und Tierschutz) Walter Dirksmeyer (Gartenbau und Diversifizierung) Henrik Ebers (Einzelbetriebliche Wirkungen) Antje Fitschen-Lischewski (Einzelbetriebliche Wirkungen) Anne Margarian (Strukturelle und regionale Wirkungen) Jan Heuer (Datenmanagement) Institut für Betriebswirtschaft...»

«MANUAL 2 Ancillary Services Manual April 2016 Version: 4.5 Effective Date: 04/28/2016 Committee Acceptance: 04/13/2016 BIC 04/12/2016 OC This document was prepared by: NYISO Auxiliary Market Operations New York Independent System Operator 10 Krey Blvd Rensselaer, NY 12144 (518) 356-6060 www.nyiso.com Disclaimer The information contained within this manual, along with the other NYISO manuals, is intended to be used for informational purposes and is subject to change. The NYISO is not responsible...»

«Spatial Perspectives: Literature and Architecture, 1850 – Present Biographies of Speakers Rosa Ainley is a writer with a background in architecture and photography and also an editor at the Architectural Association. Her most recent book is 2 Ennerdale Drive: unauthorised biography (Zer0 2012). In 2009 she was lead artist, working with muf, on Leysdown Rose-tinted, a CABE-funded regeneration ‘vision’, now in implementation. She is currently working on a book about ghost buildings: the...»

«Inhaltsverzeichnis Blogging vs. Knowledge Management 1 Einleitung 2 Kurzvorstellung der Konzepte 2.1 Knowledge Management 2.2 Blogging im Kontext von Web 2.0 3 Corporate Blogging und Verbindungen zum Knowledge Management.9 3.1 Sozialisierung 3.2 Externalisierung 3.3 Kombination 3.4 Internalisierung 3.5 Schlussfolgerung und Managementimplikationen 4 Zusammenfassung und Ausblick 5 Literaturverzeichnis Blogging vs. Knowledge Management Daniel Beverungen 1 Einleitung Wissen stellt für Unternehmen...»

«Foreman 0 DUKE UNIVERSITY Durham, North Carolina From Status to Contract: Domesticating Modernity in Wuthering Heights, The Mill on the Floss and Dracula Violeta Solonova Foreman March, 2011 Undergraduate Critical Honors Thesis Trinity College of Arts and Sciences English Department Foreman 1 ACKNOWLEDGEMENTS My deepest thanks to my thesis advisor, Professor Psomiades for her dedication, insight, positivity, encouragement, and inspiration. Also, thank you to loved ones for your constant support...»

«Untersuchungen zur relativen Chronologie der Nekropole von Marlik Christian Konrad Piller Dissertation an der Fakultät für Kulturwissenschaften der Ludwig-Maximilians-Universität München vorgelegt von Christian Konrad Piller aus Straubing München, den 7. August 2008 Erstgutachter: Prof. Dr. Stephan Kroll Zweitgutachter: Prof. Dr. Michael Roaf Tag der mündlichen Prüfung: 3. Juli 2007 Inhaltsverzeichnis Vorwort 7 1. Einleitung: Vorgeschichte und Idee zur Arbeit 8 2. Ziele und Grenzen 11...»

«©Oberösterreichischer Musealverein Gesellschaft für Landeskunde; download unter www.biologiezentrum.at BAUGESCHICHTE DER WALLFAHRTSKIRCHE VON ST. WOLFGANG IM SALZKAMMERGUT (Mit 10 Abb. auf Taf. V-X u. 3 Plänen im Text) Von Benno U l m Inhaltsübersicht : Nachrichten zum Bau der Wallfahrtskirche 63 Bisherige Forschungsergebnisse zur Baugeschichte 70 Ergebnisse der Bauuntersuchung 74 Der Bau des Langhauses 78 Der Chor 82 Die Tore des Langhauses 83 Profile und Einzelformen 85 Steinmetzzeichen...»

«STUDY TO ASSESS THE POTENTIAL IMPACT OF PROPOSED AMENDMENTS TO COUNCIL REGULATION 2299/89 WITH REGARD TO COMPUTERISED RESERVATION SYSTEMS OCTOBER 2003 Prepared for the European Commission Directorate-General for Energy and Transport by The Brattle Group* and Norton Rose** * Dorothy Robyn, James Reitzes, Boaz Moselle, Carlos Lapuerta, and Erica Carere Professor Mark Armstrong, Oxford University, academic adviser to Brattle ** John Cook and Stephen Dolan The Brattle Group, Ltd Norton Rose 15...»

«CULTURAL CENTER FOR FOREIGNERS' CALL Dialogue with an Atheist BY Dr. Manea H. Al-Hazmi General Director of the Cultural center For Foreigners’ Call Dialogue with an Atheist The Cultural Center for Foreigners' Call In the Name of Allah the Most Gracious the Most Merciful Dialogue with an Atheist Muslim: Do you believe in God? Atheist: No. Muslim: Then who created the universe? Atheist: The universe created itself or came by a chance. Muslim: According to the science of cosmology the universe...»

«Annex K6 Assessment of Effects on Residential Visual Amenity CLOCAENOG FOREST WIND FARM ASSESSMENT OF EFFECTS ON RESIDENTIAL VISUAL AMENITY Prepared for RWE Npower Renewables Ltd by Land Use Consultants August 2010 37 Otago Street Glasgow G12 8JJ Tel: 0141 334 9595 Fax: 0141 334 7789 glasgow@landuse.co.uk CONTENTS 1. Introduction Purpose and Scope of the Study Relevant Guidance 2. Assessment Methodology Introduction Definition of Visual Amenity Elements of the Development that could have...»

«Rede des Reichsführers SS bei der SS-Gruppenführertagung in Posen am 4. Oktober 1943. Zusammenfassung In seiner berühmt-berüchtigten Rede vor den SS-Gruppenführern vom 4. Oktober 1943, der so genannten Posener Rede, unternahm der Reichsführer SS, Heinrich Himmler, eine Standortbestimmung der SS im Krieg gegen die Sowjetunion und zog eine Bilanz ihrer Taten. Berüchtigt ist die Rede vor allem für die schonungslose Offenheit, mit der Himmler gegenüber seinen Generälen den Judenmord...»

<<  HOME   |    CONTACTS
2016 www.abstract.xlibx.info - Free e-library - Abstract, dissertation, book

Materials of this site are available for review, all rights belong to their respective owners.
If you do not agree with the fact that your material is placed on this site, please, email us, we will within 1-2 business days delete him.