Commit to the servers or rent them by the hour?

Ten hours of peak, fourteen hours of quiet, and two ways to pay for a server. The answer is a rule you can apply to any instance: commit to anything you need for more than about fifteen hours a day.

SaaS Intermediate Optimization free

Install free How optimization works here

After you install, this is the model to open.

Which Instance Mix Holds the Load for the Least Money?

In your spreadsheet, click the Sortia icon in the strip of icons down the right-hand edge. No strip? Click the arrow at the bottom-right to open it. You can also use Extensions, then Sortia, then Open Sortia.
Click Start from a template and put that name in the search box.
Pick the card with that name and click Load this template. It arrives on a new tab with real numbers already in it.

The answer

The cheapest mix: 4 + 6 four committed around the clock, six rented for the ten peak hours
Cost a year: $16,749 against $21,865 renting everything by the hour
The saving: $5,116 a year, a 23% cut, with the same capacity every hour
The rule it found: 14.9 hours need a box longer than that a day and committing wins

Two ways to pay for the same machine. A committed instance costs $0.238 an hour but you pay for all 24 hours whether the traffic is there or not, which is $5.71 a day. An on-demand instance costs $0.384 an hour and you only pay while it runs. The sheet opens on the plan that keeps every option open, which is buying everything on demand: ten instances for the ten-hour peak and four overnight, $59.90 a day, $21,865 a year.

Click Run and the optimizer returns $45.89 a day, $16,749 a year, which is $5,116 less over a year and a 23% cut. It commits to exactly four instances and buys six more on demand for the peak. Four is not a rounded guess. It is the off-peak floor, the smallest number that clears 1,900 requests a second overnight, and those four machines are needed every hour of every day, so they should be paid for at the committed rate.

Everything above that floor exists for ten hours and is cheaper rented. The last line of the sheet states the rule the answer obeys: a committed instance costs $5.71 a day and an on-demand hour costs $0.384, so committing wins for anything you need more than 14.9 hours a day. Commit to your floor, rent your peak, and recheck that break-even line whenever a provider changes a price.

Two runs worth doing. Set the off-peak demand to 3,000 and the committed count goes from four to six while the on-demand peak buy falls from six to four, so the commitment tracks your floor and not your peak, which is the whole idea stated as an experiment. Then set the committed rate to $0.307, a 20% discount instead of 38%: the break-even moves from 14.9 hours a day to 19.2, the day costs $52.51 instead of $45.89, and the answer does not change at all.

Four instances are needed for all 24 hours, so they clear the higher bar just as easily, and the discount would have to vanish completely before committing to your floor stopped paying. That is the reassuring version of this result: on capacity you are genuinely confident about, the commitment is not a close call. The model does not carry failover, so a serious plan adds one instance to the committed floor and calls it insurance, and it does not know that the commitment is a year long while this sheet only prices a day.

The model

It arrives on a tab called Template: Which Instance Mix Holds the Load for the Least Money?, carrying these columns:

Instances (count)
Capacity each (requests a second)
Hours a day
Cost an hour ($)
Cost a day ($)

with the model computed beside the data:

Capacity at peak (requests a second)	5,200
Capacity off peak (requests a second)	2,080
Peak headroom (requests a second)	0
Off-peak headroom (requests a second)	180
Cost a day ($)	59.9
Cost a year ($)	21,865
Hours a day above which committing wins	14.87

Once it is in your sheet

The model arrives with real numbers in it and runs as it stands, so you can press the button first and understand it second.
Change the numbers to yours. The sheet marks which cells are inputs and which hold formulas, and most labels carry a note explaining the row.
Press the run button at the bottom of the panel. It is labeled for the tool you are in, and the result lands on its own tab, with a written reading of it beside the figures.

Never used Google Sheets? Start here goes the whole way, in seven steps, and assumes nothing.

Next question

Every model like this one, and the method behind them: Optimization in Google Sheets.

Install free All 379 templates