AWS RIs vs. GCP CUDs: A Look Below The Surface
Amazon Reserved Instances (RIs) and Google Compute Engine Committed Use Discounts (CUDs) are both examples of getting a discount by committing to buy cloud services “in bulk” ahead of time. Commitments like these are an alternative to buying services on demand for just what you need.
Whether you plan to use a hybrid strategy that encompasses both clouds, or are looking to switch from one to the other, understanding RIs and CUDs starts with understanding the bigger picture of AWS EC2 and GCP GCE offerings.
In general, AWS EC2 focuses on providing depth of service. It offers a wide range of preconfigured options to anticipate any need you might have, as in “Here are a whole bunch of options for every single service we offer.”
GCP GCE tends to focus more on a balance between simplicity and customization. It offers fewer pre-configured instances, but also is incredibly customizable in that they offer custom machine types.
AWS offers about 160 EC2 instance options, with multiple RI types and attributes for each one. The type of RI you purchase needs to match the instance you’re trying to reserve in several key ways, including tenancy, platform, Availability Zone, instance type and instance size.
GCP has resource-based pricing, where each vCPU and each GB of memory on GCP is billed separately rather than as part of a single machine type. You still create instances using one of the 28 predefined machine types, but your bill reports them as individual vCPUs and memory used per hour. Pricing for custom machines is also determined by vCPUs and GB of memory, but at a higher rate. Buying a CUD gets you a discount that is applied to the aggregate number of vCPUs or memory within a region, which is not affected by changes to your instance’s machine type.
These two distinct application and pricing models require a different way of looking at the two discounts.
AWS offers Standard RIs for 1-year or 3-year terms, although shorter terms may be offered by Reserved Instance Marketplace sellers. RIs can be purchased on a No Upfront, Partial Upfront or All Upfront basis.
GCP allows you to purchase a specific amount of vCPUs and memory for a discount off of normal prices in return for committing to a usage term of 1 year or 3 years billed monthly. The discount is up to 57% for most machine types or custom machine types. The discount is up to 70% for memory-optimized machine types.
While GCP’s purchasing model is more straight-forward, AWS’ model gives companies more options to fit variable accounting models.
Key Factors for AWS RIs
Beyond making sure the RI matches your instances, there are other factors you need to integrate into your RI strategy.
Instance Size Flexibility (ISF)
With ISF, your RIs will automatically apply to all sizes of an instance family within the same region based on a normalization factor. While it is limited to instances using Amazon Linux with default tenancy, ISF helps simplify your RI planning.
When you purchase an RI, you can choose between a Standard or Convertible class. Convertible RIs can be exchanged for other Convertible RIs of an equal or higher value. Convertible RIs are a middle ground between Standard RIs and On-Demand prices that offer some additional flexibility, in exchange for giving up a bit of savings. Convertible RIs can be helpful when workloads are likely to change, or when you want to plan for potential price drops, but still get a discount of some kind.
AZ vs. Region
Amazon EC2 is hosted in multiple independent Regions world-wide, and each Region is broken down into isolated Availability Zone (AZ). RIs can either be Zonal (restricted to a specific AZ with a capacity reservation) or Regional, and can only be used for instance in their specified AZ or Region. As a side note, Regional RIs can still reserve capacity with an additional charge through Capacity Reservations. These restrictions have to be taken into account when building your RI strategy, and you’ll want to ensure your RI regions match your infrastructure.
Multiple VM OS types
Some features of RIs are only available to RIs for instances running Amazon Linux/Unix, while others may work with any Linux version, but not Windows instances. Be sure to account for these limitations in your strategy, and account for any third-party licensing fees when computing savings.
Amazon EC2 Spot Instances let you take advantage of unused EC2 capacity in the AWS cloud. Spot Instances are available at up to a 90% discount compared to On-Demand prices. The downside to Spot Instances is that you can only use them when the extra capacity is available and your workload can be interrupted. But when planned correctly, you can combine Spot Instances with On-Demand and RIs to further optimize workload cost with performance.
Key Factors for GCP CUDs
While CUDs don’t have as many factors that you need to weigh, there are some key pieces that any CUD strategy needs to integrate:
vCPUS and Memory CUDs must be purchased together
You must have both vCPUs and Memory CUDs in order to use CUDs. You can’t purchase CUDs for one without the other, so the trick is to make sure you have the right ratios. vCPUs and Memory CUDs are automatically purchased together in predefined ratios, but are applied based on the VMs used, so you want to match that ratio to your use.
Sustained Use Discounts
In addition to CUDs, Google also offers Sustained Use Discounts, which are automatic discounts that you get for running specific Compute Engine resources a significant portion of the billing month. The more you run them, the more you save. You can’t apply CUDs and SUDs to the same resource at the same time, but any resources that do not qualify for CUDs are automatically eligible for SUDs. If you’re unsure whether you’re going to use enough resources to get the most out of your CUD purchase, than you might want to weigh your potential CUD savings against SUD savings.
Preemptible VMs are similar to AWS’ Spot Instances. They are short-lived compute instances that offer savings in exchange for a lack of predictability. You can save up to 80% more with Preemptible VMs than with regular instances, but this savings comes with the potential for instance preemptions. Preemptible VMs are a great option to save on a fault-tolerant workload, which makes them a good supplement to a CUD and SUD strategy.
Bottom Line: The Same Strategy Won’t Work for RIs and CUDs
While they might seem similar, RIs and CUDs are two entirely different beasts, which means the same purchasing strategies won’t work for both. The inherent pricing and usage differences between them make an apples to apples comparison or exchange of the two impossible.
Understanding the differences and nuances of each is especially important if you’re migrating between cloud vendors or have a multi-cloud infrastructure. Your strategy needs to be built off reliable data, insights and analysis based on your actual use. Your first step should be making sure that you have the right tools — tools that will allow you to correctly allocate all of your cost and usage data so you can make the right informed decisions.
Ready to Learn More?
Download our Mastering the Fundamentals of GCP Costs e-book and take your first step toward mastering the fundamentals of GCP cost efficiency.
And to learn more about AWS RIs, read The Complete Guide to Saving with AWS Reserved Instances.