At its re:Invent conference, Amazon’s AWS cloud computing unit today announced Amazon Aurora DSQL, a new serverless, distributed SQL database that promises high availability (99.999% for multi-region availability), strong consistency, PostgreSQL compatibility, and, the company says, “4x faster reads and writes compared to other popular distributed SQL databases.”
AWS argues that Aurora DSQL will offer significantly lower latency than Google Spanner, its closest competitor.
Interestingly, AWS stresses that there is no database sharding involved here to scale the service — and that it can scale reads and writes independently. The company also highlights that Aurora DSQL will offer strong consistency, ensuring that if users opt for a multiregion approach, all regions will always show the same data at the same time.
AWS notes that to ensure resiliency, Aurora DSQL uses an active-active architecture, meaning there is always a standby server ready to take over, which guarantees “that a customer’s application is always available by enabling an application to read and write to any Aurora DSQL endpoint.” Since this is a fully managed service, AWS handles all of the security updates and manages the overall infrastructure.
To make all of this happen, the AWS team says it had to “reinvent relational database transaction processing.” Typically, ensuring strong consistency across multiple regions and having those globally distributed servers sync with what AWS describes as “microsecond accuracy” is a rather difficult feat to achieve. AWS says it can do this because it decoupled the transaction processing from the storage. Traditional methods, the company explains, would be bottlenecked by having to pass information back and forth multiple times. Aurora, however, only checks each transaction when it’s time to commit the changes and then, when the commit happens, it parallelizes those writes across regions.
One nifty twist here: To make sure that each region sees the commits in the right order, Aurora DSQL uses the Amazon Time Sync Service, which adds very precise reference clocks to every EC2 instance and synchronizes them using the atomic clocks on GPS satellites.
Having this exact time helps solve a number of data consistency issues and speeds up the processing. And while using a time-based approach may seem like a no-brainer, it takes very precise — and well-synchronized — clocks to do so. Synchronizing clocks, and doing so on a global scale, is a surprisingly difficult problem. That’s why traditionally, databases haven’t been able to use time in this way. Startups like Clockwork specialize in this. For a long time, that wasn’t actually possible, but now that these systems are available, they enable database innovations like this.
