What cost forbids, and the escapes by price
Which four things do pricing mechanisms forbid, and which escape is cheapest?
What this mechanism forbids
Four things are not allowed by the mechanisms in Pricing primitives, Design consequences, Elasticity, Control and attribution and Anti-patterns.
- Reasoning about cost from rates. The metering definition, the granularity and the utilization each move the effective price more than the rate does, and the three multiply. A "3× cheaper" engine at 20% utilization costs more than the old one at 80%.
- Treating any verdict as portable. At least four of the verdicts here do not just weaken when the price list changes. They invert at a metering boundary: compression strength (compression), the value of caching (result caching), materialization (materialization) and admission control (governors). The same design, on a different meter, gives the opposite answer.
- Assuming a performance fix is a cost fix. Concurrency hides latency. It touches no metered quantity. A thousand parallel GETs finish quickly and still cost a thousand request fees.
- Expecting the organization to discover any of this by itself. Nothing in these mechanisms emits an event. No error, no timeout, no alarm. The bill is the only signal, and it arrives late and added up.
The escapes, ordered by price
Every escape here moves a cost. It does not remove it. But the escapes differ in what they cost to use. Here they are, from cheapest to most expensive.
| Rank | Escape | What it costs |
|---|---|---|
| 1 | Change the workload (fewer queries, longer schedule intervals, fewer columns) | cheapest, but it is a negotiation with the people who generate the work, not an engineering task |
| 2 | Change the meter (commit, reserve, switch from scanned to provisioned) | a forecasting risk in exchange for a cost risk |
| 3 | Change the layout (compaction, partitioning, clustering) | a permanent maintenance obligation |
| 4 | Change the architecture (move regions, replace the engine, split storage from compute) | most expensive, and usually unreachable, because the switching cost was never budgeted |
This order is the reverse of the order in the layers of a computer system below this mechanism. There, the technical lever (a better data structure, a faster algorithm) is the cheap one. Here, the cheap lever is organizational and the expensive one is technical. This surprises engineers, because engineers reach first for the lever they know.