MODELING CONTINUOUS QUERIES OVER DATA STREAMS
In this chapter, we develop a queueing model to study the dynamics of a DSMS given a set of continuous queries over a set of data streams. In this queueing model, we need to know the input rate (or range) of each input stream. This is usually not an issue as various components of DSMSs, such as query optimizer, scheduler, load shedder, need to monitor the input characteristics of data streams for their working. Also approximate rates (or upper bounds) can be used. Therefore, it does not introduce any additional cost to use our queueing model. DSMSs only needs to monitor the input rate of an input stream periodically and the length of the periods varies as the system load changes. The rest of the chapter is organized as follows : Section 5.1 provides an overview of our queueing model. We outline the challenges in Section 5.2. In Section 5.3, we discuss how to model an individual operator in a DSMS, and provide tuple latency and memory requirement analysis of an individual operator in detail. We further extend our modeling work to an individual query plan in a general DSMS, and provide a closed-form solution to estimate QoS metrics (i.e., the overall tuple latency and memory requirement) of the output tuples of a query in Section 5.4. Section 5.6 further presents a set of our quantitative experimental results. The conclusion is presented in Section 5.7.
KeywordsService Time Schedule Strategy Busy Period Queue Size Service Discipline
Unable to display preview. Download preview PDF.