Cost
Cost is the monthly price of the hosts, parts and traffic in a design. A 1 vCPU host costs 6 USD/mo, and an 8 vCPU host costs 60 USD/mo. The best design is the cheapest one that meets every target, so each part must earn its price.
See it on the board
The same traffic runs on two boards. The simulator measured both when this page was built.
| Measure | Goal | Node API with a large worker pool | Light API that fits in 512 MB |
|---|---|---|---|
| p99 latency | under 100 ms | 10.1 s Fail | 68 ms Pass |
| Error rate | under 0.1% | 84.9% Fail | 0% Pass |
| Monthly cost | 6 USD/mo | 6 USD/mo |
Key numbers
| Number | Value |
|---|---|
| Host with 1 vCPU and 2 GB | 6 USD/mo |
| Host with 8 vCPU and 32 GB | 60 USD/mo |
| One API part | 4 USD/mo |
| CPU for one request, Node API | 1 ms |
| CPU for one request, fast API | 0.3 ms |
1,500 req/s · Node API · 3 hosts · 18 USD/mo
How it works
Every design has a bill. A host costs a fixed price each month, busy or idle. Traffic, storage and requests to managed services add costs that grow with use. The best design is not the cheapest one or the fastest one. It is the cheapest design that still meets every target.
Waste hides in three places. The first is idle capacity: a large host that runs at 10 percent. The second is a heavy runtime: a Node API uses about 1 ms of CPU for each request, and the fast API uses 0.3 ms, so the same host serves more than three times the load. The third is reserved memory: a large worker pool can claim more RAM than a small host has.
Measure first, then cut. Remove a part that does no work, choose a lighter part, or move work to a cheaper tier. Do not remove the headroom that the design needs for a failure or a spike. A cheap design that fails costs more than the money it saves.
In the levels you meet every target inside a monthly budget, and in many levels a cheaper design earns the third star.