Ilmu Komputer & AI editorial
Open AccessOA2026
Aurora DSQL: Scalable, Multi-Region OLTP
A serverless, disaggregated SQL database achieving active-active multi-region transactions with coordination only at commit time.
Marc Brooker; Marc Bowes; Mike Hershey; Zak van der Merwe; James Morle; Matthys Strydom· 2026· DOI 10.48550/arXiv.2607.13276
The core problem
Aurora DSQL addresses the challenge of providing cloud-scale transaction processing with multi-region active-active capabilities. Traditional OLTP systems struggle to balance strong consistency, ACID transactions, and continuous availability during failures while scaling elastically. DSQL is designed as a serverless SQL database that separates compute, storage, and transaction coordination into independent, horizontally scalable services. This disaggregated architecture allows each component to scale independently, enabling the system to handle millions of transactions per second and remain available during availability zone or region failures. The system executes PostgreSQL-compatible SQL, making it accessible to existing applications and developers.
Innovation
Aurora DSQL achieves elastic scaling from zero to millions of transactions per second. It provides strong consistency, ACID transactions, and continuous availability during availability zone or region failures. The disaggregated architecture allows independent scaling of compute, storage, and coordination, ensuring that bottlenecks in one component do not limit overall throughput. By deferring coordination to commit time, DSQL minimizes cross-region latency, making multi-region active-active deployments practical. The system's use of Firecracker MicroVMs for query processors enables fast startup and isolation, contributing to its serverless nature. Performance evaluations demonstrate linear scalability with added resources and low latency for both reads and writes, even under geo-distributed workloads.
Aurora DSQL addresses the challenge of providing cloud-scale transaction processing with multi-region active-active capabilities. Traditional OLTP systems struggle to balance strong consistency, ACID transactions, and continuous availability during failures while scaling elastically. DSQL is designed as a serverless SQL database that separates compute, storage, and transaction coordination into independent, horizontally scalable services. This disaggregated architecture allows each component to scale independently, enabling the system to handle millions of transactions per second and remain available during availability zone or region failures. The system executes PostgreSQL-compatible SQL, making it accessible to existing applications and developers.
DSQL's architecture is built on three key components: query processors, storage, and transaction coordination. Query processors run in Firecracker MicroVMs and execute PostgreSQL-compatible SQL without maintaining local state. This stateless design allows for rapid scaling and fault tolerance. The system employs multiversion concurrency control (MVCC) with precision timestamps to enable coordination-free reads. For writes, it uses optimistic concurrency control, deferring coordination to commit time through distributed adjudicators and the Journal replication system. This minimizes cross-region latency by requiring coordination only during commits, not for individual statements.
Why it matters
The key innovation of Aurora DSQL is its coordination-free reads and deferred write coordination, which significantly reduce cross-region latency compared to traditional synchronous replication approaches. This design leverages precision timestamps and MVCC to provide snapshot isolation without global coordination for reads. For writes, optimistic concurrency control with distributed adjudicators ensures serializability while allowing commits to proceed with minimal coordination. The Journal replication system acts as a globally consistent log, enabling adjudicators to validate transactions across regions. This architecture aligns with the taxonomy candidates of Architecture, Cybersecurity, Network, and Cryptography, as it involves distributed systems design, secure isolation via MicroVMs, network-efficient protocols, and potentially cryptographic techniques for timestamping and consistency. However, the paper does not detail specific cryptographic mechanisms, leaving room for future exploration. Overall, DSQL represents a significant advancement in cloud-scale OLTP, balancing scalability, consistency, and availability.
Who should read this
CS practitioners and researchers
Opening member content…