# Why one server stops being enough

- Source: https://intervueclub.com/course/why-one-server-stops-being-enough
- Author: IntervueClub
- Part of: HLD 101 (https://intervueclub.com/course), Thinking in systems
- Last updated: 2026-10-04

The first concept in HLD 101: how to work out which of CPU, memory, disk and network runs out first on one server, why a bigger machine is usually the first fix, and the three problems a second server brings with it.

## One server on launch day

Say you're building Instagram, and for now the whole API runs on one server: 32 cores, 128 GB of RAM and a 10 Gb/s network card. Photos come from a CDN, so the server only sends JSON. A feed request costs about 0.2 ms of CPU and returns 50 KB. On launch day traffic reaches 32,500 feed requests a second, and responses start timing out.

**Question:** CPU is the usual suspect. If you move to a 64-core machine, how many more requests a second does it serve?

**Answer:**
None. The limit is the network card. 10 Gb/s is 1.25 GB a second, and 1.25 GB ÷ 50 KB is 25,000 responses a second however many cores sit behind it. The CPU can handle 32 cores × 5,000 requests a second each (1,000 ms ÷ 0.2 ms) = 160,000 a second, so at 25,000 it is about 16% busy. On 64 cores it is about 8% busy, still serving 25,000.

![Four meters for CPU, memory, disk and network on one server as traffic climbs. Network reaches 100% at 25,000 requests a second while CPU sits at 16%, and 7,500 requests a second go unserved. The cores double to 64; CPU drops to 8% and throughput stays at 25,000.](https://intervueclub.com/animations/one-server/which-resource-runs-out.light.png)

*The network card fills first. Doubling the cores lowers CPU use and serves nothing extra.*

*The rest of this concept is free to read after signing in: https://intervueclub.com/course/why-one-server-stops-being-enough*