Hourly cloud or a monthly server? Find the break-even point
The cloud charges for the hours your machine is on, in US dollars. A server rented by the month charges all 730 hours of the month, at a fixed price in local currency. Which one is cheaper depends on one thing: how many hours your workload actually runs.
The math in four steps
- Measure the busy hours of each kind of usage in the month. Separate scheduled loads (jobs) from development and analysis, because their patterns are very different.
- Compute the cost per busy hour of each option: monthly cost ÷ busy hours.
- Find the break-even point: the number of busy hours at which the fixed monthly cost equals the hourly cost.
- Compare with the real hours. Above break-even, the monthly server pays off; below it, paying by the hour and switching off pays off.
What usually shows up
In a real case, scheduled loads ran 89% of the month: the monthly server paid off. The development environment, however, was used 25% of the month: there, paying by the hour and switching off when nobody uses it was much cheaper.
In other words, the answer is rarely "everything in the cloud" or "everything on a server". It depends on the type of workload.
Also test the cloud with a schedule
Before switching models, simulate the cloud pausing fixed capacity outside business hours. Capacity that stays on all 730 hours of the month but is only used during business hours can lose most of its cost with a simple pause-and-resume schedule.
Pitfalls
- Comparing different things: the server's full month against only the cloud's used hours, without saying so.
- Forgetting the exchange rate: cloud cost moves with the dollar; a server billed in local currency does not.
- Forgetting who operates it: your own server needs backups that are actually tested, updates and someone on call.
Want to run this math with your operation's numbers? It is the first step of our data platform FinOps.