Section: Scientific Foundations

Data Stream Management

Recent years have witnessed major research interest in data stream management systems. A data stream is a continuous and unbounded sequence of data items. There are many applications that generate streams of data including financial applications, network monitoring, telecommunication data management, sensor networks, etc. Processing a query over a data stream involves running the query continuously over the data stream and generating a new answer each time a new data item arrives. Due to the unbounded nature of data streams, it is not possible to store the data entirely in a bounded memory. This makes difficult the processing of queries that need to compare each new arriving data with past ones. A common solution to the problem of processing join queries over data streams is to execute the query over a sliding window that maintains a restricted number of recent data items. This allows queries to be executed in a finite memory and in an incremental manner by generating new answers when a new data item arrives. Due to the continuous, often very fast, arrival of new data, it is impossible to produce exact answers to queries. Therefore, approximate answers are typically provided.

In real data settings, a data stream management system may process hundreds of user queries. Therefore, for most realistic distributed streaming applications the naive solution of collecting all the data at a single site is simply not viable. Therefore, we are interested in techniques for processing continuous queries over collections of distributed data streams. An example of such queries is join queries which are very important for many applications. A streaming join computation can be useful in understanding important trends and making decisions about measurements or utilization patterns.