Using Map and Reduce for Querying Distributed XML Data

Abstract

Semi-structured information is often represented in the XML format. Although, a vast amount of appropriate databases exist that are responsible for efficiently storing semi- structured data, the vastly growing data demands larger sized databases. Even when the secondary storage is able to store the large amount of data, the execution time of complex queries increases significantly, if no suitable indexes are applicable. This situation is dramatic when short response times are an essential requirement, like in the most real-life database systems. Moreover, when storage limits are reached, the data has to be distributed to ensure availability of the complete data set. To meet this challenge this thesis presents two approaches to improve query evaluation on semi- structured and large data through parallelization. First, we analyze Hadoop and its MapReduce framework as candidate for our distributed computations and second, then we present an alternative implementation to cope with this requirements. We introduce three distribution algorithms usable for XML collections, which serve as base for our distribution to a cluster. Furthermore, we present a prototype implementation using a current open source database, named BaseX, which serves as base for our comprehensive query results.