What’s the difference between horizontal and vertical scaling?

Imagine your email platform needs to handle ten times the sending volume it handles today. You've got two options: make your existing servers bigger, or add more servers. That's the whole difference between vertical and horizontal scaling.

Vertical scaling (also called "scaling up") means upgrading a single server with more CPU, more RAM, or faster storage. It's simpler to manage because you're still working with one machine. The catch is that every server has a ceiling. At some point, you can't buy your way to more power, and if that one server goes down, everything stops.

Horizontal scaling (also called "scaling out") means adding more servers and spreading the work across all of them. It's more complex to coordinate, but there's no hard ceiling. Need to handle more load? Add another node. One server fails? The others keep going.

Email infrastructure almost always leans on horizontal scaling, and for good reason. Each email in a queue is an independent message. It doesn't care which server processes it. That makes email a natural fit for distributing work across many machines. A large ESP might run MTA clusters with dozens of sending nodes, distributed queues spread across machines, and sharded databases for subscriber data. Multiple data centers add another layer of redundancy on top of that.

But That said, it's not always one or the other. Some components inside a horizontally scaled system still scale vertically where it makes sense. A primary database server, for instance, might get a RAM upgrade before you spin up a replica. The overall architecture is horizontal, but the details are always messier than any clean diagram suggests.

If you're curious how ESPs actually put this into practice, the question on how ESPs distribute sending load gets into the mechanics.