How multiple servers serve one traffic flow
Why overload occurs
Each user action creates a request: opening a page, fetching a product list, saving a message. The server needs time to do this work. In this lesson, one server handles at most 300 requests per second. The req/s notation beside metrics means “requests per second.”
If this server receives 900 requests in a second, it completes 300. The other 600 are rejected in this model. This is overload: more work arrives than the system can finish.
Why a load balancer is needed
Adding more servers is called horizontal scaling. But the client needs to know where to send requests. In the initial diagram, the address leads only to the first server, so new servers get no work on their own.
A load balancer is a shared entry point before the servers. It accepts requests and chooses a server. One simple method is round robin: first request to the first server, second to the second, third to the third, then back to the first. Here the result is simplified to an equal split of traffic.
How to calculate required capacity
Suppose traffic is 600 requests/s. Two identical 300-capacity servers can handle it if each receives 300. One cannot cope; a third adds headroom but raises cost.
Check total capacity and actual traffic distribution separately. Even five servers cannot help if all traffic still goes to one.
What to check in the diagram
First add a server and check for an active route to it. Then add a balancer and compare each server's load, processed traffic and errors. Metrics update automatically; Check saves an attempt.
In this lesson, servers are identical and the load balancer has no capacity limit of its own. Hardware failures and a shared database are not modeled. A successful solution therefore demonstrates load distribution, not a ready-made architecture for every application.