Scalable data systems

In order to talk about scalability, we need to have a way to describe the load of a system

The load parameters of a system could be things like: requests / second, writes / reads, active users, etc.

Once we have a way to describe the load, we can investigate what happens when the load increases.

How is performance affected if we fix the resources of the system and increase a load parameter? A different question might be, given an increase of a load parameter, how much do I need to increase resources to keep the same performance?

Keep in mind that an architecture that scales well for an application is built around assumptions of which operations will be common or rare, the load parameters. If the assumptions turn out to be incorrect, the effort is wasted or counter productive.