Srovnávací analýza serverless a managed server architektur pro datovou analytiku veřejné dopravy

This thesis focuses on the comparison of serverless and managed server architectures for processing open public transit data. The main objective is to design and implement a periodic ETL pipeline for processing open data from the Golemio platform and to compare selected metrics on both architectures. A stateless containerized pipeline was designed and deployed in parallel on Azure Virtual Machines and Azure Container Apps Job. Both variants shared the same Docker image and were operated for seven days at a one-minute execution frequency against a live data source. The results demonstrated performance equivalence between the two architectures with a difference of 40 ms between the medians of total processing time. The dominant phase of each pipeline run was the database write, which accounted for up to 90 % and confirms that the workload is I/O-bound. At the chosen frequency, the FaaS variant was approximately eight times cheaper, and cold start latency in the scale-to-zero configuration averaged 11 seconds. This thesis provides empirical data and a specific break-even point for architectural decision-making in periodic I/O-bound workloads.