Addressing Edge-Cloud Microservices SLOs with Scalable Theodolite Across Multiple Kubernetes Clusters
Abstract
This deposit contains the raw measurement data, result figures, and evaluationscript behind the evaluation chapter of the Master's thesis "Addressing Edge-CloudMicroservices SLOs with Scalable Theodolite Across Multiple Kubernetes Clusters"(Ravish Kumar, Kiel University, 2026). The study benchmarks a collaborative edge-cloud image-classification pipeline thatsplits a single image stream between a Raspberry Pi 5 edge tier (ResNet-50) and anIntel Xeon cloud tier (ResNet-152), coordinated by Extended Theodolite, anSLO-driven benchmarking harness that spans two independent Kubernetes clusters. Atwo-dimensional grid of 30 configurations - six compute topologies (cloud x edgereplicas) crossed with five edge/cloud workload splits - is measured on threequality dimensions: classification accuracy, per-image inference latency, andenergy (measured on the edge via the Raspberry Pi's on-board PMIC, estimated on thecloud via the Fan et al. CPU-utilization model).