Parallel Commits is a new atomic commit protocol developed at Cockroach Labs that is capable of cutting multi-region latency in half. Nate Stewart, the Chief Product Officer at Cockroach Labs, introduces Parallel Commits to the world from the stage at ESCAPE/19 - the multi-cloud conference.
There's this idea of the hundred-millisecond rule. It's coined by the creator of Gmail. And it's a mandate that says all of your interactions, all of your client-facing or customer-facing interactions have to respond in less than a hundred milliseconds. Where did that number come from? 100 milliseconds is the threshold for a human to perceive an interaction as instantaneous. If you can deliver on the hundred-millisecond rule, you can have instantaneous applications.
Let's think about how that plays in the real world with the public cloud. Today, if you run a relational database, it's going to be running in a single region. Right? You have your big, primary machine that's handling all your SQL. And if your latency starts getting too high, you can always buy a bigger and bigger machine. What's the problem with that? Well, obviously the cost scale exponentially, but whatever we can worry about or we can lean on Moore's law to keep the cost in line. Well, what happens when your users aren't close to that data center? What happens when your users span the United States or span the world? Now you can't rely on Moore's law to increase the clock speed because now you're stuck on the speed of light. Right? You're not going to make that any faster.
In a world like that, every round trip is precious. Every time you have to go from your client to your database and back, that is something that you just... you can't avoid. So you have to think about ways to get your data physically close to your users. But one of the challenges we've had with CockroachDB is figuring out how can we support data integrity while minimizing the number of round trips. And we've made a lot of progress here over the last couple of years, but we still had too many round trips. That we still were not doing what was the theoretical minimum. And we couldn't figure out for a long time, this was really bumping into the frontiers of computer science research. But over the last six months, our team figured it out. We've figured out how to do a transaction that would ensure your data is consistent, no matter where it's accessed in a single round trip. That cuts your latency in half, that delivers great customer experiences, no matter where in the world your customers are.