Despite significant advances in data processing over the years, many data-intensive applications still run in batch mode: they require a long period of time to execute, and during that time they provide neither feedback nor control to the user. This state of affairs holds in a number of important domains, from the desktop (e.g., spreadsheets) to the back office (e.g., decision support) to new high-end applications (e.g., data mining and visualization tools). In each of these domains, batch processing is frustrating or even unacceptable for serious users. In order to investigate large data sets, users require on-line behavior: they need to get meaningful feedback while the system runs, and they need to be able to act on the feedback by controlling the system while it runs. The goal of this research project is to explore and develop techniques to make long-running, data-intensive operations have on-line functionality. This includes new research into algorithms and systems for query processing, user interfaces, data visualization, and data mining. The results of this work should allow for a much tighter loop between users and software during the process of data analysis and data exploration. This can lead to significantly enhanced power for users, and enable significant leverage of human intelligence in the data mining process. http://control.cs.berkeley.edu
