On Indexing Multidimensional Data In A P2P Context

Based on their remarkable properties (fault tolerance, scalability, decentralization), P2P systems tend to be largely accepted as a common support for deploying massively distributed data management applications. Some of the existing P2P systems are built over hybrid networks where certain peers are specialized in indexing. Others make use of DHTs as common support for data indexing. However, only a few solutions have been proposed to solve the problem of multidimensional data indexing. Many of these suffer from severe limitations regarding load balancing and scalability of the data indexed (data pieces with many dimensions are not well indexed). In this paper we propose a new approach for indexing multidimensional data using a P2P architecture. It is based on a fault tolerant and query optimal overlay network built from a skip-graph. We index data pieces of many dimensions as vectors and we adopt the polyhedral algebra to cluster values with similar properties. For a continuous distribution of values to peers, we use a hierarchical overlay network (that we call routing layer). At the end of the paper, we offer an exhaustive evaluation that compares the main properties of our architecture against similar systems.