discussion · Code of Architecture
Amazon Aurora: Design Considerations for High Throughput Cloud-Native Relational Databases
June 20, 2023 · Code of Architecture
A special Code of Architecture episode on Amazon Aurora's design: why the network becomes the bottleneck in a cloud-native database, and how separating compute from storage and treating the log as the database reduce I/O, speed up recovery, and support resilient distributed storage.
Participants
Alexander Polomodov · panelist · Technical Director, Client Interfaces, Marketing and Engagement unit · Tinkoff
Igor Maslov · panelist · Head of Core Technologies and Data Processing · Tinkoff
Vitaly Kondratov · panelist · Architect, Core Technologies · Tinkoff
Discussion focus
- 01Why does the bottleneck of a traditional database move from disks and individual nodes to the network in a cloud environment?
- 02How does separating compute from storage change database architecture and network traffic?
- 03Why does Aurora send redo records to storage and follow the principle that the log is the database?
- 04How do quorums, data segmentation, and correlated-failure analysis provide durability at cloud scale?
- 05How do log sequence numbers, asynchronous page materialization, and distributed repair reduce MTTR?
- 06Which Aurora results, constraints, and lessons matter for SaaS systems, schema evolution, and patching without downtime?