Paper 5

A DHT-based System for the Management of Loosely Structured, Multidimensional Data

Authors: Athanasia Asiki, Dimitrios Tsoumakos, and Nectarios Koziris

Volume 6 (2012)

Abstract

In this paper we present LinkedPeers, a DHT-based system designed for efficient distribution and processing of multidimensional, loosely structured data over a Peer-to-Peer overlay. Each dimension is further annotated with the use of concept hierarchies. The system design aims at incorporating two important features, namely large-scale support for partially-structured data and highperformance, distributed query processing including multiple aggregates. To enable the efficient resolution of such queries, LinkedPeers utilizes a conceptual chain of DHT rings that stores data in a hierarchy-preserving manner. Moreover, adaptive mechanisms detect dynamic changes in the query workloads and adjust the granularity of the indexing on a per node basis. The pre-computation of possible future queries is also performed during the resolution of an incoming query. Extensive experiments prove that our system is very efficient achieving over 85% precision in answering queries while minimizing communication cost and adapting its indexing to the incoming queries.