Assessment of Performance for Cloud-Native Machine Learning on Edge Devices

This article presents the performance assessment environment for conducting experiments on an edge computing platform specifically designed for machine learning. The platform consists of a virtualised environment, including a K3s cluster and Kubeflow software. The primary objective of the study was to be able to assess the efficiency of executing Kubeflow Pipelines. To verify they usability of the environment, we run simple calculus operations in a simulated parallel computing scenario. To deliver accurate performance measurements, a benchmarking environment was set up, taking into account parameters such as the number of parallel pipelines executions and nodes. The results obtained, proved the validity of the environment, by showing the impact of cluster node count on computational time. This provides valuable insights that can guide future decision-making for optimizing the performance of machine learning pipelines on edge devices.