XDB - A Novel Database Architecture for Data Analytics as a ServiceCarsten Binnig; Abdallah Salama; Erfan Zamanian; Harald Kornmayer; Sven Listing; Alexander C. Müller
In: 2014 IEEE International Congress on Big Data. IEEE International Congress on Big Data (BigData-2014), June 27 - July 2, Anchorage, AK, USA, Pages 96-103, IEEE Computer Society, 2014.
Parallel shared-nothing database systems are major platforms for efficiently analyzing large amounts of structured data. However, in order to offer SQL-like services for data analytics in the cloud, providers such as Amazon and Google do not use these systems as a basis. A major reason for this trend is that existing parallel shared-nothing database systems are expensive and that they do not fulfill many of the requirements such as elasticity and fault-tolerance needed for providing a service for data analytics in the cloud. In this paper, we present an overview of an elastic and fault-tolerant database system called XDB, which supports complex analytics. XDB builds on the following novel concepts: (1) a partitioning scheme that supports elasticity with regard to data and queries, (2) a cost-based fault-tolerance scheme that allows to recover from mid-query faults, and (3) adaptive parallelization techniques to better support complex analytical queries. XDB is implemented using a middleware approach on top of multiple nodes each hosting an instance of a single node database system (MySQL in our prototype). Initial experiments show that our novel concepts effectively support elasticity, fault-tolerance and complex analytics when compared to the traditional behavior of existing databases.