A common misconception about Reserved Instances

While working with engineering teams there is a common misconception about how Reserved Instances or Savings Plans work that keeps coming up in conversations.

Many engineers think of reservations as a "bag of hours" that you can consume when and how you need. For example, a typical 365 year has 8760 hours in it, and when you buy a 1 year reservation on a, lets say, c6i.2xlarge AWS instance your are indeed buying 8760 hours of use of such an instance.

For this example, the on-demand pricing at the time of this writing is $0.3648 per hour (Ireland region). With a 1 year reservation and all-upfront payment you get a much better price of $0.26017 per hour.

While the engineer focuses on the 8760 hours (the quantity), they are missing the point that reservations are a price optimization strategy. They think that they can buy those hours now, not use them for two months and then use them all up in three months.

If they would do such a thing, they would be in for a surprise:

1. They would pay $2,279.09 first in the up-front payment.
2. They would then leave 9 months (2 at the start plus 7 at the end) worth of that amount unused ($1,709.32 to the waste bin).
3. They would use 2,184 hours during those 3 months that would qualify and be covered by the reservation.
4. And they would use another 6,576 hours NOT covered by the reservation and paid at the on-demand price, for a total of $2,398.93 more.
5. All in all, $4,678.02 spent. That is $1,482.37 more than the $3,195.65 they would have paid at on-demand rates.

To be fair, they would probably notice that something is amiss when they receive the first $799.64 bill. And hopefully put the paid computing resources to some use.

As the late Chester Bennington would sing "The clock takes life away...".

Reservations are a commitment to use some computing capacity for a period of time, and while you don't need to use it fully to get some juicy savings, to maximize the benefits of a reservation you must be using it from the very moment you purchase it to the very end of the purchased period.

For each and every hour during that time, the cloud provider will check your usage for a match with your reservation and apply it if it finds one. Not necessarily the same instance each time, as instances can (and should in many cases) be ephemeral.

If no resource matches, that hour is lost forever.

Matching resources exceeding the amount of reservations you made are billed at on-demand rates.

Key takeaways

1. Reservations (and Savings Plans) are a price lever, a commitment to use computing resources for a period of time.

1. Always purchase reservations on compute capacity you are already using and plan on using for at least over two thirds of the reservation term (as a rule of thumb). Never on what you plan to use in the future.