AWS Revamps Pricing in Bid to Reduce Risk and Improve Flexibility
Amazon Web Services (AWS) has added a new discount model to give customers a quicker on-ramp to savings while locking them into multiple-year term agreements. The latest discount plans are based on annual or multiple-year spend commitments on the AWS platform.
AWS Pricing Models
Discount plans from AWS are nothing new. AWS pioneered the concept and commercialized cloud services as being truly elastic, scalable, and, most of all, inexpensive. Traditionally this has taken the form of the AWS Reserved Instance (RI) model. In the RI model, AWS customers select a specific instance and its geographical Region and Activity Zone and lock that capacity up for either a one-year or three-year term at a locked-in discount up to 75% off the On-Demand rate.
Later on, as competition from Microsoft and Google introduced their own, more flexible versions of reserved instances, AWS added a Convertible Reserved Instance which allows for more flexibility for customers to re-size their environments within the region by “modifying their instance family, OS, or tenancy with their RI.”
Specific exclusions apply for AWS GovCloud (US) and AWS China (Beijing) but plan to be added at a later date. Convertible RIs can still yield up to a 54% discount but average ~45%, according to AWS.
The RI Challenge
RI plans have been instrumental over the past eight years in helping to propel AWS to a whopping ~$36 billion annual run rate for cloud infrastructure-as-a-service (IaaS) revenues. Even with Microsoft’s Azure growing at yearly rates north of 50-60%, they are nowhere near approaching the scale of Amazon in pure-play IaaS.
With RI plans, AWS customers must determine which specific Elastic Cloud Compute (EC2) cloud instances are required inclusive of the AWS Region and Activity Zone locations for those instances. Recent changes now allow for RI migration across Activity Zones within the same region. However, RIs must be selected for either a one-year or three-year term commitment. For organizations to minimize the risk of over- or underbuying and to maximize capacity utilization, a rigorous cloud architecture must be completed with a high degree of accuracy.
Many may argue that developing an overall cloud architecture and consumption plan is best practice and should occur in most circumstances anyway. This is true. Alas, the world of IT is not such a static environment where well-made plans of yesterday necessarily hold true today. Changing customer demands, shifts in business models, and the rapid rate of change may quickly make yesterday’s well-defined cloud architecture woefully out of date. Then what can you do with all those heavily discounted RIs you purchased?
By the way, calculating cloud IaaS capacity is no trivial task. In one study by TSO Logic, they examined usage patterns for organizations seeking to migrate on-premise workloads to a public cloud environment. Interestingly, 84% of the on-premise workloads were generally overprovisioned, and if migrated in a lift-and-shift manner to the cloud, overall costs would increase by 20%. If the cloud workloads were optimized for those migrated workloads, costs could decrease by up to 35%. The insight uncovered by this study is that “on an apples-to-apples basis, the cloud is more expensive.”
This is only one example but serves to illustrate the pitfalls that await the designers of cloud-hosted environments. Other factors such as fluctuating demand patterns, seasonality, management of non-production instances, and New Product Introduction (NPI) are a few more. Another firm estimates that up to $14.1 billion in IaaS compute spend is wasted, comprising 53% of IaaS spend on compute!
Back to the RI conundrum, where planning for a cloud environment, even when the requirements are stable and well-known, takes a tremendous amount of time. Just review the RI pricing page at AWS to get a sense of this challenge. In short, it takes a lot of work to design the most cost-effective RI service mix even when all other characteristics of the IaaS environment are reasonably well-known.
AWS Savings Plans
AWS’s new discount model, called Savings Plans, removes much of the complexity of RI plans. There is a trade-off in the form of a more in-depth “vendor lock,” as annual or term revenue commitments are more significant in scale than with the RI model.
Source: SoftwareReviews, November 19, 2019
There are two new plans for AWS customers to consider:
Compute Savings Plans
The more flexible of the two new Savings Plans, the Compute Savings Plan allows customers to reserve any instance family in any region. An instance family is comprised of the combination of compute, memory, and storage capabilities. This discount will max out at 66%.
EC2 Instance Savings Plans
The EC2 Savings Plan offers a tad higher discount potential, at 72%, but imposes more restrictions on the capacity commitment. The EC2 Savings Plan is limited to a specific instance family.
- AWS now offers three types of rates when selecting an instance: On-Demand, Reserved Instance, and Savings Plan.
- Spend commitments are made on a per hour basis, and discounts apply on all compute usage up that limit. Afterward, the rate defaults to the On-Demand price.
- The Savings Plan commitment is assessed hourly, and AWS will apply the discount to the consumption area with the deepest discount.
A basic comparison table of AWS Savings Plans:
|
Compute Savings Plan |
EC2 Instance Savings Plan |
|
|
Discount (vs. On-Demand) |
Up to 66% |
Up to 72% |
|
Flexible Across: |
Region, Family, OS, Tenancy |
Size, OS, Tenancy |
|
Applies to: |
EC2 or Fargate |
EC2 |
Savings Plan can be purchased in one-year or three-year terms with All Upfront, Partial Upfront, or No Upfront payment terms.
Our Take
The introduction of AWS Savings Plans may be a game-changer for AWS customers. Those who have struggled to adequately allocate resources to proper cloud environment workload plans and architectures and those who continue missing the mark on forecasting actual usage vs. planned usage may find Savings Plans a true “saving grace.”
It is an arduous task to accurately forecast the usage in cloud environments, as there are multiple factors to consider, and many of the inputs are best guesses as opposed to hard facts. As such, many organizations regularly underestimate the monthly costs of their cloud environments. AWS Savings Plans provide the following improvements over the legacy RI model:
- Flexibility: Savings Plan allows Instance migration across regions, which is still not permitted in RI plans. This is a significant improvement in flexibility for large global organizations that operate in multiple AWS regions.
- Auto-discounting: AWS has improved the flexibility of RIs over the years, including the advent of Convertible and Size Flexible Reservations. However, customers must take conscious action to continually evaluate and adjust these RIs to maximize the discount potential. Under Savings Plans, AWS performs an hourly evaluation and applies the Savings Plan where it yields the deepest discount.
- Product Expansion: Historically, discount plans have only applied to EC2 instances. Savings Plans applies to Fargate as well, opening up the possibility of new discount plans that may focus on Database, Analytics, Machine Learning, or others.
- Purchase Method(a): Savings Plans are purchased by the hour, not by the instance. This forces a mindset shift as customers need to think about total spend vs. instance capacity in isolation.
- Purchase Method(b): AWS allows for several methods of payment from All Upfront to options requiring minimal upfront spend. The savings rate does not change (unlike with RIs), allowing customers to stretch capital expense budgets or to spend quickly in “use it or lose it” end of budget cycle scenarios.
Oh, and if you already purchase under the RI model, don’t fret. If usage qualifies for both plans, AWS will first apply the Reservation spend and apply the Savings Plan spend in other areas of consumption.