Performance vs Scalability
The one-sentence test
A service is scalable when it can handle more load by adding resources without an unacceptable loss of performance. Near-proportional gains are the ideal, not part of the definition, and shared bottlenecks determine how close you get.
So there are two independent problems, and confusing them wastes a lot of engineering time:
A performance problem is visible even at low load: the work itself is slow because of an algorithm, query, dependency, or round trip. Replicas usually do not shorten that critical path.
A scalability problem appears as load grows: requests contend for a finite resource. More resources can help if the bottleneck is one you can replicate or partition. The low-load/high-load comparison is a diagnostic, not a complete definition.
Slow with one user → performance problem Slow only under load → scalability problem Doubling servers helps the second Doubling servers does not the first
Before you scale anything, check whether the system is slow when it is idle. If it is, you are about to buy N copies of a problem instead of fixing it.
6 components7 connections0:00
Recording…