Vertical scaling, also known as scaling up, includes adding more power to your existing server. You keep the same database node, but you give it more CPU, memory, or storage to do its job.

This is often seen as the traditional way to scale relational databases. It keeps the architecture simple because all your data still lives in one place.

Benefits

• Simplicity: Your code stays the same because the database still looks like a single machine to your application
• Easier maintenance: You only have one server to manage and update
• Consistency: You don't have to worry about syncing data between multiple servers

Challenges

• Hard limits: Every machine has a maximum capacity. Eventually, you cannot add more hardware to a single server
• Downtime: You often need to take the database offline for a moment to perform these hardware upgrades
• Single point of failure: If that one server goes down, your entire application can go down with it

Managed database services, such as Cloud SQL, can help make vertical scaling easier by letting you change your machine size with just a few clicks, but it’s important to remember you may still hit those physical hardware limits.

Small-scale: Imagine a fast-growing startup that runs a web app on a single database. When traffic starts to climb, they can move to a larger instance, which in turn gives them more compute power to handle the extra visitors without needing to rewrite their application code.

Enterprise-scale: A large enterprise with a legacy system might use vertical scaling too. These systems typically rely on complex relationships between data. Because their code is built for a single database, they can scale up by adding more power to that one machine instead of trying to move to a distributed system.

Internal needs: Some companies also use vertical scaling for internal reporting. They may add more RAM to their primary database server so it can process large, resource-heavy queries faster without slowing down the experience for regular users.

Horizontal scaling, or scaling out, indicates adding more machines to your database system instead of making one machine more powerful, which distributes the work across a group of servers. It is a core concept in modern cloud computing and distributed systems.

Engineers use techniques like sharding, where you split your data into smaller chunks across different servers, and replication, where you copy data to several nodes, to make this work.

Benefits

• Limitless growth: You can keep adding more machines as your needs grow
• High availability: Because you have multiple servers, the system stays online even if one node fails
• Cost-efficiency: You can use cheaper, standard hardware instead of buying one massive, expensive server

Challenges

• Complexity: Your application needs to know how to talk to multiple servers, which adds design challenges
• Data consistency: Keeping data in sync across many different machines can be difficult to manage
• Code changes: You might need to change your application code to handle a distributed database environment

A distributed SQL database like Spanner can help you manage these challenges. Spanner handles the hard work of syncing data across multiple machines for you, so you do not have to split your data manually or lose the safety of a relational database. This allows you to scale out without making your system too complex.

Capacity: A global e-commerce platform might use horizontal scaling during big sales events. When millions of people visit the site at once, the system can automatically rebalance data across new nodes to handle the traffic, ensuring that no single server gets overwhelmed.

Redundancy: Mission-critical financial applications sometimes also use horizontal scaling. They can distribute their database across multiple physical locations, known as availability zones. This setup helps ensure that if one data center has an issue, the application stays up, keeping their promise of data consistency.

Proximity: A global gaming company might deploy database nodes in different parts of the world. By putting a database node closer to the player, they can provide a fast, smooth experience with very low lag. For example, if a player lives in Tokyo, their profile data is stored on a Tokyo database node for lightning-fast reads and writes.

The main difference between horizontal and vertical scaling is how they add resources to your system. Vertical scaling increases power to one machine, while horizontal scaling adds more machines to your network.

When deciding on how you want to scale your database, the choice often depends on your budget, how much data you have, and how much uptime you need. The typical advice is as follows:

• If you have a predictable, moderate amount of traffic, it’s easier to start with vertical scaling because it’s simpler and fast to set up.
• If you’re building for global scale or need constant uptime, it’s recommended you plan for horizontal scaling from the beginning. It takes more work to design, but it can save you from major headaches later on.