Skip to content
Learn/

The Application Layer

1 / 4

What separation actually buys

Separating the web tier from the application tier — and the application tier into services — lets each piece scale on its own axis. If image processing is CPU-bound and the API is IO-bound, running them together means you buy CPU for machines that mostly wait on the network.

It also lets teams deploy independently, which for a large organisation is often the real motivation. The architecture diagram is a picture of the org chart more often than anyone admits.

What it costs is that every call between services is now a network call. Local code can still be slow or throw, but it shares a process fate and cannot produce an ambiguous network timeout. Across a service boundary latency, partial failure, and uncertain outcomes are routine, so every caller needs a timeout and an explicit retry or fallback policy.

The failure mode is a distributed monolith: services that must all be deployed together and cannot function independently. You have paid the full cost of distribution and kept none of the independence.

6 components7 connections0:00

Traffic
20Kreq/s
p50
48ms
p99
95.8ms
Errors
0.09%
Dropped
18req/s
Cost
$624/mo