The Aurora DSQL team has published a comprehensive research paper on Arxiv covering the full architecture of Aurora DSQL, AWS's scalable multi-region OLTP database. The paper details query processing, transactions, replication, and the control plane in one place. Key highlights include the disaggregated architecture that separates replication from transaction isolation and storage, the use of Optimistic Concurrency Control (OCC) to prevent client blocking and avoid metastable failures, strongly consistent scalable reads, and pushing scan operations down to storage near local SSDs. The paper also argues that advances in time distribution, datacenter networking, and distributed protocols have made strong consistency the right default for cloud-scale systems, overturning earlier assumptions that eventual consistency was necessary for availability and latency goals.
