• data stream processing
• middleware
• real-time thread
• scheduling

Recent years have seen a growth in the volume of available data, often in the form of streams, produced by various sources (social networks, IoT devices and so on). One of the objectives of the data stream processing paradigm is to analyse this volume of data in order to extract information, alerts and knowledge. Whereas in the past stream processing was carried out on centralised platforms such as clouds or clusters of workstations, the trend today is to perform parts of the analysis of streaming data close to those who produce it, i.e., at the edge of highly distributed computing systems where resources are limited. Because of the limited resources available, a scheduling of these applications that takes into account specific metrics within the application itself ensures a gain in performance. The purpose of Lachesis, a middleware developed by the Parallel Programming Models research group of the University of Pisa and the Distributed Systems group at the Chalmers University of Technology, is to help the operating system scheduler to better schedule data stream processing applications. Building on what had been done with Lachesis in the past, in this thesis we explore new optimisations and kernel mechanisms that allow Lachesis to apply a finer and more effective scheduling of running applications, by taking advantage of mechanisms offered by the operating system that had never been tested before. The experiments, conducted on an Edge computing device, showed improvements in both throughput and latency over both the original version of Lachesis and the OS scheduler.