Cloud OLTP Architectures and Cost Optimization Framework

This technical report presents an analytical model framework for optimizing cloud OLTP (Online Transaction Processing) systems. It systematically addresses key questions regarding the best architecture for cloud OLTP systems, the cost implications of running specific workloads, and the optimal provisioning of hardware instances. The report outlines six OLTP architectures, including Classic, In-Memory, Remote Block Device (RBD), High Availability Disaster Recovery (HADR), Amazon Aurora, and Microsoft Socrates, detailing their trade-offs in terms of durability, availability, throughput, latency, and cost. The authors develop a model that incorporates various workload requirements, such as dataset size and performance metrics, to determine the most cost-effective architecture and hardware configuration. The findings are based on a calibration with the LeanStore storage engine and real-world pricing from AWS. The report aims to guide database administrators and engineers in making informed decisions regarding system design and resource allocation for cloud environments.