Section: New Results

Embedded Data Management

Participants : Nicolas Anciaux, Luc Bouganim, Lionel Le Folgoc, Yanli Guo, Saliha Lallali, Philippe Pucheral, Iulian Sandu Popa, Shaoyi Yin.

Inspired by low cost economic models, this work draws the idea of a one-dollar database machine, with the objective to disseminate databases everywhere, up to the lightest smart objects. In contrast to traditional database machines relying on massively parallel architectures, the one-dollar database machine considers the cheapest form of computer available today: a microcontroller equipped with GBs size (external) Flash storage. Designing such a database machine is very challenging due to a combination of conflicting RAM and NAND Flash constraints. To tackle this challenge, this work proposes a new paradigm based on database serialization (managing all database structures in a pure sequential way) and stratification (restructuring them into strata when a scalability limit is reached).We show that a complete DBMS engine can be designed according to this paradigm and demonstrate the effectiveness of the approach through a performance evaluation. This work capitalizes on previous results related to the indexing of Flash resident data [16] and has also obvious connections with the more general study we are conducting on Flash-based data management (see Section 6.2 ). Partial elements of this solution have been demonstrated at [13] . In 2012, we have extended our previous results on indexation of flash resident data [IS] and we have proposed the design of a complete DBMS engine [DAPD] complying by nature with the conflicting RAM and NAND Flash constraints we are facing. Currently, we work at the extension of the embedded DBMS engine to support document data (e.g., text documents or any type of documents that are tagged)) and spatio-temporal data (e.g., vehicle trajectory data or any type of time-stamped and/or geo-located data).