How can I use rolling Savings Plans to reduce commitment risk?
Savings Plans offer discounts on compute of up to 72% compared to on-demand pricing in exchange for usage commitment (in $/hour) for a one or three-year period.
Savings Plans are more flexible than most commitment-for-discount offerings. The most flexible type of Savings Plan applies across the core compute services (Amazon EC2, AWS Fargate, and AWS Lambda) and across EC2 instance size, operating system, tenancy, availability zone, and region. This flexibility accommodates continuously evolving workloads and avoids unused commitment.
There are three types of Savings Plans, each available with multiple commitment terms and payment options. This article uses a one-year compute Savings Plan with the partial upfront payment option for reference and calculations, but the concepts are equally applicable to other combinations.
Drawbacks of a single monolithic Savings Plan
A common misconception is that one must or should cover their target commitment level in a single annually renewed Savings Plan. This approach constitutes a single large commitment, presenting an especially unappealing risk to organizations as they reassess exposure to unpredictable swings in business conditions, such as those caused by the recent pandemic.
It also encourages building 6+ month usage predictions into the commitment level (i.e., hitting a moving target), which is daunting and difficult, at best. Even normal fluctuations in usage can result in waste from over or under committing.
As a result, an organization may underuse Savings Plans and their discount benefit due to risk aversion and the difficulties inherent in predictions.
Alternative: rolling Savings Plans
An organization can purchase Savings Plans at any time and at any desired level of commitment – concurrent active Savings Plans are additive.
This article describes a technique for overlapping periodic (ex: quarterly) Savings Plans that reduces commitment risk, increases discount coverage, and relieves the burden of long-range usage predictions. For example, by distributing your target commitment level over four staggered annual Savings Plans, you gain the ability every quarter to reduce or eliminate up to ~25% of your commitment, if business conditions warrant (i.e., a “lever to pull”). And, each quarterly renewal consideration won’t need to try and incorporate usage predications farther out than ~3 months.
About commitment risk
Commitment-for-discount offerings are great for the common (and desirable!) case of business growth and, therefore, usage. However, there is a risk that usage drops unexpectedly and severely. If it drops well-below the commitment level, it would erode the pay-as-you-go value of cloud computing since expenses would no longer align with usage.
Concern over commitment risk is legitimate and front-of-mind for organizations hit by downturns in business due to the recent pandemic or other unforeseeable circumstances. Organizations are exploring, and in some cases, requiring “levers to pull” to align expenses with falling revenue, if needed.
Reducing commitment risk
The rolling Savings Plan approach replaces a monolithic Savings Plan – the latter is deconstructed into multiple, approximately equal, staggered Savings Plans (ex: quarterly), each of which has a commitment level set based on recent and short-term future usage considerations.
Each individual Savings Plan still has a term of one year, but collectively the organization has a Savings Plan expiring every 3 months that represents ~25% of the total commitment (for quarterly purchases). Therefore, the organization could drop its commitment level up to ~25% on very short order, if necessary. And in catastrophic business circumstances, it could drop by another ~25% every quarter going forward – as opposed to waiting up to one year to reduce the commitment level at all.
Each quarterly re-evaluation point represents a “lever to pull” per the needs of the business.
Increasing discount coverage
Each quarterly re-evaluation point is also an opportunity to adjust the commitment level for normal fluctuations in usage. For example, to increase according to business growth or pull-back as needed. As a result, the commitment level more closely tracks usage, and discount coverage is improved compared to annual re-evaluation.
Avoiding long-range planning
More frequent re-evaluation points also preclude projecting usage out farther than 3 months because another re-evaluation point will occur at that time. Removing the motivation for attempting to incorporate future predicted growth into a Savings Plan purchase also reduces the risk of over-commitment.
Here’s an example of quarterly rolling Savings Plans, starting from zero existing Savings Plans:
I determine my target total commitment level to be $10.00/hour, based on the traditional manner of analysis provided by the AWS Cost Management Console, and my understanding of how recent usage compares to expected usage for the next 3 months.
I purchase my first one-year term Savings Plan at 1/4 of my target commitment level:
SP 1: ($10.00/4) = $2.50/hour
One quarter later, I re-evaluate my target total commitment level. Perhaps due to natural business growth it has risen to $10.40/hour. I now purchase my second Savings Plan at 1/3 of the unmet target commitment level:
SP 2: ($10.40 – $2.50)/3 = $2.63/hour
I repeat this exercise at the end of the second quarter for my third Savings Plan purchase. This example assumes there has been more natural business growth to $10.82/hour:
SP 3: ($10.82 – $2.50 – $2.63)/2 = $2.84/hour
Finally, at the end of the third quarter, the total commitment level has now dropped to $10.49/hour:
SP 4: ($10.49 – $2.50 – $2.63 – $2.84) = $2.52/hour
For the fourth quarter of my first year, I have four active Savings Plans totaling $2.50 + $2.63 + $2.84 + $2.52 = $10.49/hour. Collectively, these plans cover my entire target commitment level as of the start of my fourth quarter.
At the one-year mark, my first ($2.50/hour) Savings Plan expires and I can effectively reset it based on current business projections. Hopefully, I will get to set it higher due to strong business growth. But, if necessary, and as demonstrated in Figure 3 below, I can reduce it (or even let it expire) without renewal.
Notes on Figure 3: each Savings Plan’s commitment level is based on the target commitment at the start of its quarter and the undulations are exaggerated to illustrate the example and concepts more clearly. The target commitment level for the graph is 90% of the actual usage, representing a bullish projection of the business, yet providing some safeguard against over-committing. Note how even when usage drops (in Q3/Q4 and again in Q6/Q7) the flexibility in the rolling Savings Plans prevents over-commitment.
After the initial ramp-up (Q1 – Q3), the discount coverage (the colored areas for the Savings Plans) more closely matches the target commitment over time than a monolithic Savings Plan would, due to the more frequent, quarterly re-evaluation points.
The Savings Plans become unbalanced following periods of abrupt changes in target commitment, but still retain their benefits compared to a monolithic Savings Plan. For example, the Savings Plan starting in Q5 (SP 5) is smaller than others because of the Q3/Q4 drop in usage.
Drawbacks of rolling Savings Plans
Adopting or migrating to rolling Savings Plans requires ramping up more slowly to one’s target commitment level compared to the immediate purchase of a single monolithic Savings Plan. Therefore, the discount benefit over the first 9 months (for our quarterly example) is lower. This can be seen by comparing the early discount coverage of Figure 3 to Figure 4.
However, from that point forward and in perpetuity, the discount benefit is maximized, the commitment risk is reduced, and long-range usage predictions are eliminated. Over the course of many years, the overall impact of the reduced discount for the first 9 months diminishes.
One could oversize their early (ex: SP 1 and SP 2) rolling Savings Plan purchases to reduce this drawback, but then subsequent rolling Savings Plan purchases (ex: SP3 and SP 4) would be undersized, resulting in weaker “levers to pull”, if it became necessary when they expire. There is a trade-off between early discount coverage and later commitment risk reduction.
Rolling Savings Plans also require more frequent evaluations (ex: quarterly instead of annually). But since the evaluations do not need to incorporate long-range usage predictions they are much simpler. The net effect will vary by organization, but is generally not significant.
Rollover frequency and Savings Plan terms
While this article uses quarterly rollover frequency to illustrate the technique, of course there are other choices. For example, you could use three overlapping Savings Plans with a re-evaluation every 4 months, or two overlapping Savings Plans with a re-evaluation every 6 months. A higher rollover frequency gives you more flexibility to adjust commitments, but also requires more re-evaluations. In theory, you could have 12 overlapping Savings Plans with monthly re-evaluation, but it seems there is a point of diminishing returns on effort vs. flexibility.
On a related note, there’s the Savings Plan term. Savings Plans are available for either one- or three-year periods. When using a three-year term Savings Plans, many more Savings Plans (12) are required to maintain the flexibility of quarterly re-evaluation points.
Savings Plans offer considerable savings on compute compared to on-demand pricing and are flexible to accommodate continuously evolving workloads, but are sometimes underused due to commitment risk or the imposition of long-range usage predictions.
The rolling Savings Plan technique reduces commitment risk, increases discount coverage, and relieves the burden of long-range usage predictions by distributing the target total commitment level over multiple staggered (ex: quarterly) annual Savings Plans. This gives you more adjustment points at which you can change the commitment level either a little, to optimize discount for typical usage fluctuations, or a lot (ex: ~25%) to counteract severe, unexpected business conditions.