Why did you add that box?

Every box must answer: what symptom did we measure, what does this fix, and what new problem does it add. Browser → backend → database is a complete architecture for a lot of products. The first real symptom is usually CPU saturation or p99 latency on the single backend under load; the fix is a second instance behind a load balancer, and the new problems are session state and deploys across instances (so the service must be stateless).

Next symptom: the database is read-heavy and its CPU is the bottleneck — say 90% reads, the same 1,000 products fetched repeatedly. A cache in front of it cuts DB load, and adds staleness, invalidation and stampedes. A queue arrives when a request does work the user does not need to wait for (sending email, generating a PDF) or when bursts exceed what synchronous processing can absorb; the cost is eventual consistency and an invisible backlog. A CDN is justified by static bytes or geographic latency (150 ms round trip to a far origin), and adds purge and cache-key problems.

The point is not to avoid these components; it is that each rung of the scaling ladder buys capacity by adding a problem, and adding the problem before the capacity is needed is pure cost.