Simplify your AWS Reserved Instance capacity planningErik Carlin, Co-Founder and Chief Product Officer
Capacity planning with RIs is a nightmare... "complicated" if we're being generous. If you've used or experimented with Standard RIs historically, your RI planning activity has probably looked "bottoms-up", looking at each EC2 instance based on your understanding of the expected life of that instance and making an RI decision accordingly.
If this is you, you're not alone. We generally see the primary decision criteria to use an RI (or not) oriented around the expected life of the individual EC2 instance. Planning this way requires a high level of precision about the future of a particular instance: "Will we be using this instance 1 or 3 years from now? Will we need to use something different based on how we plan to refactor our app? Will AWS release a new instance type that is better suited for us?" If you are that good at future forecasting, get into investment management ASAP.
Forecasting RIs instance-by-instance is not only complicated but highly time-consuming if you have a moderate to large EC2 footprint. Not surprisingly, we see many users that only apply RIs to the bare minimum parts of their footprint and pay on-demand rates for everything else. This radically reduces the discount they achieve on their overall AWS bill.
However, the complexity and manual effort is not the worst part about forecasting RIs "bottoms-up"... it's that you are making the RI decision(s) based on the wrong criteria. You are focused on the expected life of an instance when you should focus on which RIs maximize your Effective Savings Rate (i.e., the discount off the on-demand price).
To simplify capacity planning and maximize your Effective Savings Rate, we suggest forecasting "top-down". At ProsperOps we are able to RI capacity plan for our customers "top-down" for a variety of reasons:
- We use Convertible RIs, instead of Standard RIs, which are infinitely flexible over the course of the RI term commitment (i.e., they can be exchanged to apply to different EC2 instance families, tenancies, operating systems, and more as the EC2 environment changes).
- We make RI commitments based on the aggregate spend expected in a region versus the expected life of individual EC2 instances.
- We think about term based on the commitment to the platform (i.e., "do I expect to be spending money on EC2 in this region over the next 12-36 months?") versus the instance ("will I be using a t2.medium for the next 12 or 36 months?").
- We algorithmically execute the RI purchase and exchange process to optimize for the highest Effective Savings Rate, instead of which EC2 instance we predict will be alive for the entirety of the RI term commitment.
By capacity planning "top-down", we distill the planning activity down to two macro variables:
- "What amount of your expected aggregate spend are you comfortable committing to in the region?"
- "How long or short are you committed to the cloud platform (taking into account a mix of 12 and 36 month term commitments, as well as breakeven economics)?"
With these two variables, we can reduce the planning activity down to 30 minutes, while accommodating anticipated future risks and retaining the optionality to increase commitments at any point in the future. To learn more about ending your RI capacity planning nightmare, please check out our site. If you have any questions, you can reach out to us there or email us at email@example.com.