Instance size flexibility for EC2 Reserved Instances
Reserved EC2 instances (RIs) can be used to reserve instance compute capacity for a fixed term, in exchange for a sizeable discount.
I have previously written about reserving RDS instances. This post focuses on EC2 reservations, and specifically, how to use AWS’ normalization factor to make things easier and more beneficial.
What is the Normalization Factor
The Normalization Factor (NF) can be thought of as a measure of compute usage per hour, but only within the context of billing.
Looking at the table below, if you are using a t2.nano instance for one hour, then you have consumed 0.25 units.
| Instance Size | Normalization Factor |
|---|---|
| nano | 0.25 |
| micro | 0.5 |
| small | 1 |
| medium | 2 |
| large | 4 |
| xlarge | 8 |
| 2xlarge | 16 |
| 4xlarge | 32 |
| 8xlarge | 64 |
| 10xlarge | 80 |
| 16xlarge | 128 |
| 32xlarge | 256 |
Where this comes in to play, is when your reservation discounts are being applied to the account.
The older mechanism
Previously, your RI must exactly match an instance for a discount to be applied. In the above example, if you had purchased a reservation of a single t2.small instance, this would not have matched your usage of the t2.nano, so no discount would be applied.
Reservation
| Instance Type | Instance Family |
|---|---|
| t2.small | t2 |
Usage
| Instance Type | Instance Family |
|---|---|
| t2.nano | t2 |
This billing mechanism led to wastage across RIs and required a lot of upfront planning on how your infrastructure requirements may change over the course of your resevation term.
The newer, more flexible, mechanism
Using the newer mechanism, your usage is taken in terms of compute units, and is applied to all sizes of instances within an instance family. Back to our above example. Your usage of the t2.nano instance would have clocked up 0.25 units. A portion of your reservation belonging to your t2.small instance can be applied to your t2.nano usage.
Keeping it in the family
As mentioned above, the discounts can be applied to all sizes within a family. Amazon also allows you to combine instances.
By reserving a single t2.small instance, we have an NF of 1 to play with. This could be split between two t2.nano instances and one t2.micro. This makes for much easier forecasting of cost and usage across your infrastructure layer.
Caveats
- This flexibility only applies to Linux instances.
- NFs cannot be shared across families.
- Instances must be using shared tenancy (the default).