Section: Scientific Foundations
Ubiquitous data management
The vision of the future dataspace, a physical space enhanced with digital information made available through large-scale networks of smart objects is paint in [43] . The management of data in such dataspace differs dramatically from the mainframe database setting. In this context, the data sources are moving, managed by highly constrained computing devices, might get temporarily or permanently disconnected and have at best a partial knowledge about their environment.
This setting strongly impacts the way data is managed locally. Actually, not only data but also data management techniques (e.g., querying, access control, transaction) must usually be embedded in highly constrained hardware devices. For example, sensor networks collecting weather or pollution data [38] are evolving towards real distributed databases in which each sensor acts as an active node (i.e., as a micro-data server queryable remotely) [44] . Protecting the confidentiality of portable folders (e.g., healthcare folders, users' profiles) is another motivation to embed data management techniques into tamper-resistant devices (e.g., smart cards) [10] . Embedded database techniques are also required in every context where computations have to be performed in a disconnected mode. To conceive embedded database components is however not obvious. Each target architecture is specifically designed to meet desirable properties (portability, energy consumption, tamper resistance, production cost, etc), under imposed hardware constraints (maximum silicon die size, memory technology, etc), to tackle specific application's requirements. The challenge is then twofold: (i) being able to design dedicated embedded database components and (ii) being able to set up co-design rules helping hardware manufacturers calibrating their future platforms to match the requirements of data driven applications. While a large body of work has been conducted on data management techniques for high-end servers (storage, indexing and query optimization models minimizing the I/O bottleneck, parallel DBMS, main memory DBMS, replication and fault tolerance, etc), few research effort has been placed so far on embedded database techniques. Light versions of popular DBMS have been designed for powerful handheld devices but DBMS vendors never addressed the more complex problem of embedding database components into chips. Recent works have been conducted on smart card databases and on data management techniques for sensor networks but this research field is still at a preliminary stage.
The dataspace setting also impacts the way queries are expressed (spatio-temporal conditions, continuous queries) and executed (decentralized control, scarce local computing resources, uncertain availability of the data sources). Distributed query management has been extensively studied for thirty years [45] , considering a reduced collection of data sources managed by high-end servers. These methods are irrelevant in a context involving potentially millions of data sources managed by lightweight devices. Query management in Peer-to-Peer systems and in Data Grids address the scalability issue and the unpredictable availability of data sources but do not consider lightweight devices. The first works to consider distributed queries (restricted to filters and aggregations) over lightweight devices have been conducted in the sensor network field. Hence, regular queries distributed over a large collection of full-fledged databases managed by lightweight devices remains an open issue.
