Metrics and Tool for Evaluating Data Stream Processing Systems
André Leon Sampaio Gradvohl · 2018
With the Internet of Things and many other data sources continuously generating events, service providers who analyze these data need fast and flexible software. These software, called data stream processing systems, can be created from scratch or, as is more often the case today, use frameworks that already implement many of the functionality required for stream processing. In fact, programmers implement these software as distributed systems today because they need frameworks to deal with huge, fast and high throughput streams. That is the reason why a centralized system would hardly be able to handle data streams without eventually discarding some events. Therefore, it is important that, before putting a data stream processing system in production, we can simulate the scenarios that this system will handle and analyze its behavior. In this sense, this paper presents the general characteristics of the data stream processing systems, as well as some of the main metrics for the analysis of this sort of system. In addition, we present a new tool called B2-4DaSP that facilitates the task of analyzing data stream processing systems. We also illustrate some examples of the B2-4DaSP use for the analysis of two simple data stream applications.