The research of the database systems group has focused on data stream management, scientific data management, index structures, query language design, and query optimization and processing.
Data stream management
Data streams appear in many emerging applications, such as financial monitoring, sensor networks, RFID systems, etc. Online filtering, aggregating of the data streams as well as detecting complex events over them is crucial to these applications. The group has been focused on the scalability problems in processing data streams and developed several algorithms for memory management, data stream dissemination, query load distribution as well as multi-query optimizations in a distributed and dynamic environment.
Scientific sensor data management
Environmental monitoring data collected from wireless sensors typically needs to be further interpreted by using external models before being utilized for scientific research. This is not only because raw sensor data are noisy and incomplete, but also because there is often a mismatch between what scientists desire from the data and what raw sensor data can offer. This phase of data processing is called data preparation, which is often interlaced tightly with the query processing phase that comes next and hence should not be treated separately. The group has built a data processing system, which integrates both the data preparation and query processing phases and offers an easy-to-use and efficient data processing service to support environmental scientists' routine sensor data manipulation tasks. By borrowing the wisdom from relational DBMS, the system adopts a generic data model, a generic query processing framework as well as a generic processing optimizer to provide an efficient and integrated data processing platform.
In-network sensor query processing
Wireless sensor networks are increasingly being deployed in many important applications to enable users to query the physical world, such as environmental monitoring, health care monitoring, military surveillance, traffic monitoring, etc. To ease the deployment of such applications, the wireless sensor network could be abstracted as a database. With this logical abstraction, users can issue declarative queries without having to worry about how the data are generated, processed, and transferred within the network, as well as how the sensor nodes are (re)programmed. Query optimization techniques can also be applied to optimize the network operations. When there are multiple queries posed to the resource constrained wireless sensor network, it is critical to leverage the sharing of their resource consumption within the network. The group has developed several multi-query optimization algorithms to achieve this goal.
When non-sequential access to data is required in a database system, indexes are created. The group has worked on data structures with relaxed balance as a tree-based index implementation technique. For efficiency in data retrieval, tree structures should be balanced. However, balance can be costly to maintain when updating tree-based structures. The relaxed balancing techniques offer flexibility in the scheduling of balancing operations which can be utilized in numerous ways to improve over-all performance. At the same time, it is possible to establish amortized worst case guarantees for the rebalancing matching the classic results where updating and rebalancing are tightly coupled.
The group is cooperating with members from many international institutions, including Ecole Polytechnique Fédérale de Lausanne, ETH Zürich, the National University of Singapore, and the University of Toronto.