The four things a vendor can meter
Why is a payload-independent path metered per event?
There are only four kinds of meter. Which one a vendor picks is not arbitrary. A vendor meters the resource that is scarce for the vendor, that cannot be shared at the same moment by two customers (it is rivalrous), and that is cheap to count.
| Meter | Dimension | Scarce resource being rationed | What you control |
|---|---|---|---|
| Stock | $/GB·month | occupied byte-time | size × retention |
| Occupancy | $/vCPU·hour, $/warehouse·minute | a machine unavailable to others | utilization |
| Flow | $/GB transferred | link and topology capacity | placement |
| Event | $/request, $/TB scanned | a path whose cost does not depend on payload | request count, layout |
The fourth row is the one that is not obvious, and it explains per-request pricing completely:
A 1-byte GET and a 1-GB GET use the same amount of front-end processing, authentication, metadata lookup and connection handling. If the vendor billed only per byte, small requests would be nearly free and could overload the service. The per-request meter is a rationing mechanism first and a revenue source second.
Where energy and carbon are priced or reported, they form a parallel meter with different gradients from dollars: a cheaper region may use more carbon per kWh, so the two bills do not track each other.