Today’s post dives into the architecture of Aurora DSQL, focusing on SQL execution and transactional reads. One interesting aspect is DSQL’s ability to scale compute, read/write throughput, and storage space independently by separating storage and compute. DSQL utilizes a service-oriented architecture and Firecracker MicroVMs to enable dynamic compute scalability. Snapshot isolated transactions in DSQL ensure consistent views by leveraging multiversion concurrency control. Avoiding caching and coherence bottlenecks, DSQL allows storage to carry out filtering and aggregation tasks, reducing round-trips between the query processor and storage. This disaggregated approach aims for independent horizontal scalability of critical database components. Tomorrow’s post will explore DSQL’s write path to complete the big picture.
https://brooker.co.za/blog/2024/12/04/inside-dsql.html