Fig. 6

The overall architecture of START. The Web-based user interface helps users select signal tracks, construct STQL queries, submit queries, and retrieve execution results. It also provides various additional functionality, such as user management, storage for queries, data files and result files, and sharing of queries with other users. The metastore provides information about the stored signal tracks in the backend database. When a query is sent to the backend system, it is handled by a driver that consists of three main components. First, a compiler checks for potential syntactic and permission errors, and produces a parse tree of the query if no errors are found. Second, an optimizer analyzes the parse tree and determines an execution plan optimized for efficiency. Third, an executor calls the underlying system to execute the query. The underlying system is based on the Hadoop framework, which distributes the data files needed and performs the actual computations on multiple machines in parallel. When a job is finished, the results are stored and the user is notified to preview or download them using the user inferface