Introducing Blocking Operators in DCEP-Sim

Smart cities, health care, weather monitoring, and agro-industrial and environmental are a few of the expected Internet of Things applications. These systems contain large numbers of sensors that continuously produce data based on their surroundings. Many applications of this data require quick and correct decision-making. Datastream processing systems and especially Complex Event Processing (CEP) systems are designed for such applications. CEP systems regard the data produced by sensors as digital representations of real-world events. CEP systems are often distributed to analyze the sensor data with minimal latency. Evaluating the performance of distributed CEP systems is challenging because it is hard to control all the parameters that affect performance. The purpose of DCEP-Sim is to simulate distributed CEP networks accurately, thus relieving experimenters from managing real distributed CEP systems. Furthermore, it aims at having a CEP engine model that can represent any existing CEP engine. This goal is challenging because there is no standard for CEP; different CEP engines produce different outputs based on syntactically equal queries. Currently, DCEP-Sim is in an early stage of development. One of its deficiencies is support for blocking operators, which are operators that produce results based on more than one event. Aggregate operators are examples of blocking operators because they compute single values from bounded sequences of events. In this thesis, we have designed and implemented blocking operators for DCEP-Sim. This task was solved by assigning each blocking operator a window, a mechanism that splits the stream into finite chunks called window extents. The blocking operators perform their calculations on the window extents when they have reached their maximum number of events. The users define window size and how much the window extents should overlap. In addition, they may group the events in a window extent by one or more attributes. This solution aims at being general enough to model any CEP engine. The implementation was tested, and the performance was evaluated in diverse experiments. It proved to be a scalable and robust system, giving accurate results and being able to run for weeks. Furthermore, both runtime and memory consumption increased reasonably with the increment of the essential parameters. For example, an experiment proved that a CEP engine with 1000 operators can process 10 000 events in 40 seconds