This chapter is from the book
This chapter is from the book
This chapter is from the book
The Concept of Reserved Instances
Did you ever get an offer to purchase something ahead of need at a good discount, something like a vacation timeshare or even a cemetery plot? Reserved instances offered by public cloud providers are the same. They are compute configurations and/or storage instances that are purchased in advance of actual need, and the provider offers a discount to pay in advance and reserve the purchases for your use.
There are a few things to discuss here, good and bad. First, let’s cover the good. If you’re good at planning ahead and understand exactly when and how you will need a compute or storage instance from a cloud provider, this is a good option to consider. However, based on my experiences, chances are you’re overconfident about your planning abilities, most especially if this is your first cloud project. But let’s assume that some of you have this superpower, and then the reserved instances may be a way to get more for less.
If you have not guessed yet, the bad around reserved instances is that most of the people planning for the use of cloud resources in the future do not have enough experience or luck to make the right choices. When they stock up on reserved instances, most overpurchase. Unlike that giant cooler the big box store took back a month after you bought it, all sales are final with cloud providers.
Figure 3-7 depicts what happens most often. Instances are purchased from the cloud provider for a predetermined cost, which is now a sunk cost. The instances typically expire after a specific amount of time, usually one to three years, and end up being largely unused. Thus, you may see a cost layout as shown in Table 3-1.
TABLE 3-1 This is an example of the business case for leveraging reserved instances: a refund.
Resources and Outcomes |
Cost |
|---|---|
10 reserved instances, x86 CPUs, 6 GB of memory, running Linux, usable for one year. |
$20,000 one-time fee. |
Number of instances used, 4, with the remainder returned to the cloud provider. |
Considering what we paid for the reserved instances, we say that we paid $5,000 per instance. |
Cost of purchasing the same instances as needed. |
$3,000 per instance. |
Overpaid/underpaid. |
$20,000 – $12,000, or $8,000 overpaid for the same cloud resources. |
)
FIGURE 3-7 Although reserved instances may make sense for a small few, most overbuy and underutilize reserved instances that typically cannot be returned for a refund.
So, do reserved instances ever make sense? Sure, such as when you’re relatively certain that your instance need will be X and then you use X. However, as we covered previously, most of those who do cloud resource and cost planning still don’t fully understand what will be leveraged and in what amount of time.
Another scenario that would make sense is if you forecast the need for 100 compute instances over the next year. You could purchase 50 reserved instances to reduce the cost. If you fall a bit short, such as needing only 63 instances over the year, you’re still ahead of the game.
In this case, you paid for 50 instances at a discount that you needed and used. With 50 instances discounted, you spent less money than if all the instances were purchased at full price as needed, even though you were off by 13 instances. Although you did not correctly predict the number of instances you planned to leverage, the strategic use of reserved instance pricing allowed you to save some money without much risk of letting the prepurchased instances go to waste. Now that more FinOps teams and tools are starting to emerge, we’re seeing more enterprises take advantage of this strategy.
Enterprise Discount Program (EDP) is a cost-savings approach that most enterprises don’t understand. AWS is the cloud provider known for this program, but other public cloud providers have something similar. The idea is that organizations that commit to a predetermined annual cloud spend (typically $1 million or more) can obtain automatic discounts. The discounts will vary based on spend commitment. These are functionally like reserved instances but provide more holistic cost advantages for larger enterprises that will be spending a great deal on public cloud services.
The general recommendation here is that you should avoid using reserved instances unless or until your cloud resource planning is so good and accurate that you feel it would be an advantage. For most of us, it’s a gamble. Much like those subscription services you sign up for and used once or twice, they rarely deliver the value you expect.