Part 2 of 8 · 1 min read
Where cost comes from
Common practice, not from the books.
A cloud bill is like an electricity bill: you pay for what is on, whether or not you use it. Most of it comes from four sources (common practice):
| Driver | What you pay for | Typical trap |
|---|---|---|
| Compute | Machine time: instances, containers, functions | Machines left running at low load, or sized for a peak that is rare |
| Storage | Gigabytes kept per month, by speed class | Old data, copies and backups never deleted |
| Network | Data moved out of the cloud, between regions and between zones | Chatty services spread across regions |
| Managed services | Databases, queues and tools charged per use or per size | Over-provisioned capacity nobody checks |
Observability adds its own cost. Module 3 showed that full logging of a busy service can reach hundreds of gigabytes a day, and UDS says storing and processing all that is not cheap.
A made-up bill. A service runs 20 instances at 0.10 per hour: 20 × 0.10 × 730 hours = 1,460 per month. It keeps 50 TB of logs at 0.02 per GB-month: 50,000 × 0.02 = 1,000. It sends 30 TB out at 0.08 per GB: 30,000 × 0.08 = 2,400. Total about 4,860. The biggest line is network, which most people would not have guessed.
Cost driver
A cost driver is a quantity that makes the bill grow, such as machine hours, gigabytes stored or gigabytes moved.
Example: if monthly traffic doubles and each request sends the same bytes, the network line also doubles.
Check yourself: 1. Which four sources make most of a bill? 2. In the example, which line is largest? 3. How does heavy logging raise the bill?