Architecture and methods for flexible content management in peer-to-peer systems

Abstract

The operation of dedicated content repositories is a change in perspective of content lifecycle management: their application largely promises both technical and financial benefits. Today, centrally managed static client-server architectures are the prevailing design approach for content repositories. However, systems built according to this paradigm inherently lack flexibility regarding the support of different content models and functional properties (for example, dynamic reconfiguration) as well as non-functional aspects (for instance, scalability). A decentralised approach based on the peer-to-peer architecture paradigm is proposed in this thesis to overcome these drawbacks. Peer-to-peer architectures promise a more flexible architecture pattern migrating into more and more application domains. In spite of the fact it has been nearly a decade that popular peer-to-peer systems appeared as an auspicious paradigm for distributed computing, successful operation is still associated with basic file sharing applications; most of these (monolithic) systems miss sophisticated data management features for concurrent usage-as required by content repository systems. In this thesis, the applicability of the peer-to-peer paradigm for the implementation of content repository functions is investigated, and an architecture and methods to enable flexible content management in peer-to-peer systems are presented. Research challenges originate in terms of (i) reflecting different characteristics and relationships of content, (ii) supporting an adequate content repository model-both at functional and non-functional level, for example, to ensure reliability and consistency properties, and (iii) coping with peculiarities of a heterogeneous, dynamic peer-to-peer environment. This thesis has created the following main research contributions: A generic and modular architecture for peer-to-peer based content repositories is presented: this thesis describes and analyses the logical view of a content repository to benefit the provision of a modular system design. The introduced architecture is able to abstract from specific data management details and to support local and remote storage areas. A particularity concerns the definition of a persistent storage layer representing the connection to integrate peer-to-peer based methods for enabling flexible content repository functions. A concept is given to enable the generic mapping of (fragmented or distributed) conten to persistent storage: the concept supports the integration of flexible storage policies to decide which degree of flexibility in content management is desired; it shows a way to annotate (content) items and to map these to corresponding back-end storage entities (resources). DhtFlex is introduced as a method to enable flexible, atomic data management for structured peer-to-peer overlays: it enables the implementation of flexible content repository functions in such overlays. DhtFlex is developed as a modular component to ensure the consistency of distributed replicated data resources in the face of concurrent updates. It enables flexible and efficient data operations using its concept of annotated data resources: on the one hand, DhtFlex provides atomic operations on replicated mutable data resources; on the other hand, DhtFlex is able to handle immutable data resources in a special way in order to support more efficient processing for them. Reconfigurable group communication is shown as a method to support flexible replication management for hybrid peer-to-peer overlays: it constitutes a major building block to implement the concept of reconfigurable peer-to-peer service groups, which is suggested to extend the approach to flexible content repository functions so that it can also be used for hybrid peer-to-peer overlays. For example, peer-to-peer service groups may be used as indexing peers to consistently administrate the replicated metadata of a repository. A generic method is stated to facilitate a decentralised and dynamic code loading of services for peer-to-peer overlays: the approach benefits integration and maintenance of a system being operated by many peers. The method represents a major building block of peer-to-peer service groups. The definition of a generic peer architecture enables the mechanisms to extend a peer's abilities dynamically at runtime. As a result, this thesis evaluates how the proposed architecture and the peer-to-peer based methods enable flexible content management: on the one hand, the architecture is evaluated using qualitative considerations, for example, considering its suitability in the context of both a cross-enterprise business collaboration scenario, and an intra-enterprise knowledge management scenario. On the other hand, the methods-on the basis of DhtFlex and peer-to-peer service groups-are analysed by quantitative observations respecting reliability, consistency, reconfigurability, scalability, and performance properties. The introduced solutions narrow the tradeoff between requirements of content repositories and inherent properties of peer-to-peer systems. © Vieweg+Teubner | GWV Fachverlage GmbH, Wiesbaden 2010. All rights reserved.