If you read what Amazon has to say about AWS Reserved Instances vs. On Demand instances, making a choice between the two pricing options “for applications that have steady state or predictable usage” is a no-brainer—or so it would appear.
Although committing to a level of service—and paying for it up to three years in advance—will generate a significant discount, the likelihood is the discount will not quite reach the giddy heights of “up to 72%” indicated on Amazon’s pricing pages.
As this look at AWS Reserved Instances vs. On Demand instances will demonstrate, the actual discount you’ll receive could be far less than implied. Furthermore, due to the introduction of AWS Savings Plans in 2019, there may be more effective ways in which you can take advantage of AWS´ discount programs.
The difference between Reserved Instances and On Demand
In terms of compute options and configurations, Reserved Instances and On Demand instances are the same. The only difference between the two is that a Reserved Instance is one you rent (“reserve”) for a fixed duration, and in return you receive a discount on the base price of an On Demand instance.
There are two classes of Reserved Instances:
- Standard Reserved Instances lock you into using one instance family, on the same operating system.
- Convertible Reserved Instances offer the flexibility to change families, operating systems and tenancy, but at the cost of a slightly smaller discount.
To best demonstrate the cost differences between AWS Reserved Instances vs. On Demand instances, we’ll be focusing solely on Standard Reserved Instances. The same principles apply to Convertible Reserved Instances and Scheduled Reserved Instances, only the savings you’ll be able to make when compared to On Demand instances will be slightly smaller.
How Reserved Instance pricing is calculated
AWS gives you three options on how to pay for Reserved Instances, with each reflecting the amount of discount you’re eligible for. Using an m6g.12xlarge instance in the US East region as an example, these are the discounts you can expect to receive compared to the standard On Demand pricing ($1.8480 per hour) as of September 2020:
Standard One Year Term
| Payment Option | Upfront | Monthly | Effective Hourly | Discount |
|---|---|---|---|---|
| No Upfront | $0 | $846.22 | 1.1590 | 37% |
| Partial Upfront | $4,835 | $402.96 | 1.1040 | 40% |
| All Upfront | $9,478 | $0 | $1.0820 | 41% |
Standard Three Year Term
| Payment Option | Upfront | Monthly | Effective Hourly | Discount |
|---|---|---|---|---|
| No Upfront | $0 | $586.41 | $0.8030 | 57% |
| Partial Upfront | $9,774 | $271.49 | $0.7440 | 60% |
| All Upfront | $18,375 | $0 | $0.6990 | 62% |
As a side note, one of the reasons AWS appears to be head and shoulders above the competition is that many of its larger customers take advantage of the All Upfront option to maximize savings. These forward payments are included in AWS’ quarterly revenue statements as “deferred” or “unearned” revenues. As the pre-paid contracts end, and businesses adopt cloud containers or a multicloud policy, expect AWS´ apparent growth rate to decline and other cloud providers to encroach on its market share.
The issue with AWS On Demand vs. Reserved Instance pricing
The issue with AWS On Demand vs. Reserved Instance Pricing is that Standard Reserved Instance pricing is fixed. If prices are reduced during the life of a one or three year contract (as they have been year-on-year during the past decade), customers see no benefit from the price reduction and are committed to paying the Reserved Instance price for the duration of the contract.
Convertible Reserved Instances are different inasmuch as customers can exchange their reservations for different instances, even outside the same family, to take advantage of price drops and performance increases. AWS Savings Plans are even more flexible since customers commit to a specific spend, rather than to a specific service.
Using Moore’s Law and historical evidence as a guide—and considering future price reductions as cheaper competitors extend their presence in the market—it’s reasonable to anticipate occasional price reductions of up to 25%. Using the m6g.12xlarge instance as an example, the following illustrates how such price reductions can erode the benefits of Standard Reserved Instance pricing:
Standard One Year Term
| Payment Option | Amount Paid | On Demand Cost | Saving | “Real” Discount |
|---|---|---|---|---|
| No Upfront | $10,154 | $12,141 | $1,978 | 17% |
| Partial Upfront | $9,670 | $12,141 | $2,471 | 20% |
| All Upfront | $9,478 | $12,141 | $2,663 | 22% |
Standard Three Year Term
| Payment Option | Amount Paid | On Demand Cost | Saving | “Real” Discount |
|---|---|---|---|---|
| No Upfront | $21,110 | $28,076 | $6,966 | 25% |
| Partial Upfront | $19,548 | $28,076 | $8,618 | 31% |
| All Upfront | $18,375 | $28,076 | $9,701 | 35% |
This table makes two assumptions—firstly that a 25% price reductions occurs the day after a one or three year term is contracted, and secondly that the m6g.12xlarge instance is fully utilized for the duration of the contract. Neither of these assumptions are likely events and probably balance each other out (although under-utilization of resources is the more likely of the two assumptions). With this additional information, it becomes more clear that Convertible Reservations are a more logical choice than Standard ones.
AWS Reserved Instances vs. On Demand: conclusion
Although the above table indicates there are still advantages to AWS Standard Reserved Instances vs. On Demand instances, these only apply when certain conditions are met—i.e. the instances where Reserved Instance pricing is applied are correctly provisioned and running 24/7. Any over-provisioning or under-utilization will result in unnecessary “cloud waste.”
Over-provisioning can be addressed before committing to a Reserved Instance contract by taking advantage of cloud cost management software. If there’s the likelihood of under-utilization over the duration of the contract, the best option is to opt for Convertible Reservations over Standard for instances that are running in production. For instances not in production, another proven cost-saving technique is scheduling stop/start times, which can save you up to 65%.
Alternatively, if you’d like to know more about AWS Savings Plans, read our ebook about utilizing Savings Plans here.
