Dynamic Auto-Scaling Solutions for DICOM Image Import in Azure : Adaptive Resource Allocation for Fluctuating Medical Imaging Workloads

Efficient and reliable handling of fluctuating medical imaging workloads is critical in modern healthcare systems. This thesis explores the performance of reactive autoscaling for DICOM image import in Kubernetes environments, with a focus on evaluating and extending the Horizontal Pod Autoscaler. We simulate realistic hospital workload patterns which consists of ramp-up, ramp-down, and burst scenarios. We compare scaling behaviors across both default CPU-based metrics and custom metrics such as active DICOM associations. The simulations are run in a cloud-based Azure Kubernetes Service cluster, using a benchmarking setup that sends DICOM images via the DCMTK protocol suite.Our results show that metric selection and scaling policy configuration significantly affect responsiveness, stability, and resource efficiency. We also identify practical challenges, such as load balancer limitations and the time it takes to provision new nodes during scaleup events, both of which impact system responsiveness and reliability. The findings highlight both the potential and limitations of reactive scaling and suggest directions for future improvements, including more intelligent metric tuning, load-aware routing, and proactive or hybrid scaling approaches tailored to healthcare environments.