Browsing by Author "Weiler, Simon"

Now showing 1 - 2 of 2

Open Access
Automatic resource scaling in cloud applications - case study in cooperation with AEB SE
(2021) Weiler, Simon
As an increasing number of applications continue to migrate into the cloud, the implementation of automatic scaling for computing resources to meet service-level objectives in a dynamic load environment is becoming a common challenge for software developers. To research how this problem can be tackled in practice, a state-of-the-art auto-scaling solution was developed and implemented in cooperation with AEB SE as a part of their application migration to a new Kubernetes cluster. Requirement elicitation was done via interviews with their development and IT operations staff, who put a strong focus on fast response times for automated requests as the main performance goal, with CPU, memory and response times being the most commonly used performance indicators for their systems. Using the collected knowledge, a scaling architecture was developed using their existing performance monitoring tools and Kubernetes' own Horizontal Pod Autoscaler, with a special adapter used for communicating the metrics between the two components. The system was tested on a deployment of AEB's test product using three different scaling approaches, using CPU utilization, JVM Memory usage and response time quantiles respectively. Evaluation results show that scaling approaches based on CPU utilization and memory usage are highly dependent on the type of requests and the implementation of the tested application, while response time-based scaling provides a more aggregated view on system performance and also reflects the actions of the scaler in its metrics. Overall though, the resulting performance was mostly the same for all scaling approaches, showing that the described architecture works in practice, but a more elaborate evaluation on a larger scale in a more optimized cluster would be needed to clearly distinguish between performances of different scaling strategies in a production environment.
Open Access
Human-AI collaboration for immersive analysis of spatiotemporal ensemble data
(2024) Weiler, Simon
Many simulations and experiments produce large amounts of spatiotemporal data, for example consisting of sets of two-dimensional positional recordings over a long time interval. The high dimensionality of the data, together with its complex time-dependent behaviors, greatly limits the possibilities of manual analysis using traditional tooling. This thesis presents a novel approach to the visual analysis of spatiotemporal ensemble data by combining an immersive and intuitive virtual reality (VR) interface with interactive machine learning elements. By defining queries for specific spatiotemporal patterns, users are able to arrange the entire ensemble in a three-dimensional workspace based on the similarity between members, while individual members and their temporal behavior can be examined in detail using an intuitive three-dimensional visualization utilizing space-time cubes. Through a small-scale user study, the workflow and VR implementation have been tested on their usability, together with a comparison between different interaction techniques in terms of task efficiency and user experience. Results show that even users with little VR experience responded positively to the three-dimensional interactions and intuitive data exploration, while also achieving high ratings in immersion and engagement, despite an initial learning curve and some visual clarity issues.