Will AWS Recommendation Engines ever recommend RIs that would be utilized less than 100% (based on historical data)?

Question

I understand that AWS's underlying RI and SP recommendation engine(s) uses historical data to (imperfectly) make recommendations to fit projected future usage.

We know, mathematically, that purchasing just enough RIs to ensure 100% utilization does not necessarily maximize savings, nor does purchasing to ensure 100% coverage.   The AWS documentation states that the engines will attempt to "maximize your estimated savings"  (https://docs.aws.amazon.com/cost-management/latest/userguide/ri-recommendations.html).    Is it fair to assume, then, that the engine will recommend that customers buy (some) RIs that won't be fully utilized?    Do we state that anywhere?   Do we explicitly state in any of our guidance that it might be OPTIMAL in some circumstances for customers to plan and implement RIs with less than 100% utilization?

Accepted Answer

It's correct - the engine may in fact recommend RIs or SPs purchase that will get you to less than 100% utilization overall. RI/SP Utilization isn't a metric you should be going for, when it comes to business KPIs, and when it comes to deciding how much to commit. The actual realized savings is a much better indicator.

I can't confirm if we have this anywhere in the documentation, but I can certainly confirm that this is how recommendations engine works, after helping customers understand their commitment recommendations for many years. 100% utilization DOES NOT mean maximum savings. It all depends where's the break even point for your specific usage patterns.

How recommendations engine works? Think about it this way - it makes incremental assumptions... for example, first it checks, how much you'd be savings if you reserved $1 commitment... then how much you'd be savings if you reserved $2. It will reach certain point when it will make no sense to commit more (let's say... it might be at $10/hour) - in which case the engine will provide the most optimal recommendation (let's say, it could be $9.999/hour commitment); and it's true that the projected utilization may be below 100% let's say if you have spiky workloads, but if you would still benefit from committing to SPs.

Answer

Hey Jim!
I understand where you are coming from with this and that would be why I lean towards recommending Savings Plans at 80% coverage based on current usage. Once an end user understands their future use cases they can then apply another layer of an SP to move closer to 100%. If you still want more information I would suggest reaching out to an Optics Specialist or account concierge.

Will AWS Recommendation Engines ever recommend RIs that would be utilized less than 100% (based on historical data)?

Relevant content