Networks for Grid Applications. Third International ICST Conference, GridNets 2009, Athens, Greece, September 8-9, 2009, Revised Selected Papers

Research Article

Percolation-Based Replica Discovery in Peer-to-Peer Grid Infrastructures

Download412 downloads
  • @INPROCEEDINGS{10.1007/978-3-642-11733-6_6,
        author={Francesco Palmieri},
        title={Percolation-Based Replica Discovery in Peer-to-Peer Grid Infrastructures},
        proceedings={Networks for Grid Applications. Third International ICST Conference, GridNets 2009, Athens, Greece, September 8-9, 2009, Revised Selected Papers},
        proceedings_a={GRIDNETS},
        year={2012},
        month={6},
        keywords={P2P Grids content search percolation},
        doi={10.1007/978-3-642-11733-6_6}
    }
    
  • Francesco Palmieri
    Year: 2012
    Percolation-Based Replica Discovery in Peer-to-Peer Grid Infrastructures
    GRIDNETS
    Springer
    DOI: 10.1007/978-3-642-11733-6_6
Francesco Palmieri1,*
  • 1: Federico II University of Napoli
*Contact email: francesco.palmieri@unina.it

Abstract

Peer-to-peer Grids are collaborative distributed computing/data processing systems, characterized by large scale, heterogeneity, lack of central control, unreliable components and frequent dynamic changes in both topology and configuration. In such systems, it is desirable to maintain and make widely accessible timely and up-to-date information about shared resources available to the active participants. Accordingly we introduce a scalable searching framework for locating and retrieving dataset replica information in random unstructured peer-to-peer Grids built on the Internet, based on a widely known uniform caching and searching algorithm. Such algorithm is based on bond percolation, a mathematical phase transition model well suited for random walk searches in random power law networks, which automatically shields low connectivity nodes from traffic and reduces total traffic to scale sub-linearly with network size. The proposed schema is able to find the requested information reliably end efficiently, even if every node in the network starts with a unique different set of contents as a shared resources.