AWS Working to Scale Aurora DB Writes Globally, Like Spanner
AWS is currently working to add global-scale transaction processing technology to its Postgres-compatible Aurora database service, which would give the likes of Google Cloud’s Spanner and Cockroach Labs a run for their money, an executive with the company said at this week’s re:Invent conference.
According to Jeff Carter, vice president of relational databases for AWS, the company is currently developing the capability for customers to scale database writes globally in Amazon Aurora, its hugely popular managed relational database services that’s compatible with PostgreSQL and MySQL.
“There are two aspects” to scaling globally, Carter told Datanami in an interview at re:Invent this week. “Scaling read and scaling write. On scaling read, I think we have that covered. On the scaling write side, we have projects in place today that we’re not announcing today, but we are working hard on getting delivered to help with the scale-of-write aspect.”
AWS has the scaling read side covered with DynamoDB, its speedy key-value store. Following its historic migration away from the Oracle database that was completed in 2019, Amazon.com is now a big user of DynamoDB to serve data involved with its huge ecommerce operation. Carter was involved with that migration project for years before joining the AWS side of the house. About 95% of the 7,500 Oracle database instances that Amazon.com was running are now running on Aurora, while the remainder of the workloads (the most read-heavy workloads) went to DynamoDB.
As a NoSQL database, DynamoDB makes tradeoffs to achieve that global-scale database read capability. But scaling writes is a tougher problem to solve, largely because the database writes often are financial transactions that demand a high degree of fidelity and compliance with ACID (atomicity, consistency, isolation, and durability).
In other words, enterprise are not willing to make the same tradeoffs when it comes to writes, which is why they continue to demand the ACID database capabilities that remain associated with relational databases. That means better engineering is required to maintain ACID compliance when transactions are arriving haphazardly across global networks with large and unpredictable latencies.
Google Cloud addresses the challenge with Spanner by using atomic clocks and GPS receivers to ensure the order and synchronicity of transactions, while Cockroach Labs uses the RAFT consensus algorithm and a multi-version consensus control (MVCC) system into its database.
Cockroach Labs’ database is wire-protocol compliant with Postgres. And earlier this fall, Google Cloud announced that Spanner now features a PostgreSQL interface. According to AWS, interest in Amazon Aurora is booming. In fact, it’s the fastest growing product in the history of AWS. Why the sudden massive interest in PostgreSQL? “It works,” said Carter, who spent 30 years building databases at Teradata before joining Amazon.
Carter didn’t say anything more about the project, other than to say development work is underway. “We’re working to get it there on the Aurora camp,” he said. “No announcement today. But stay tuned.”