Welcome to this lesson on maximizing AWS savings with Reserved Instances and Savings Plans.

Both of these models allow you to save money by committing to use AWS services over a specific period of time, either a one- or three-year term. But they have some important differences in terms of how they get applied and how much flexibility they offer.

So first let's talk about Reserved Instances, or RIs. RIs allow you to reserve a specific instance configuration, which consists of an instance family, size, operating system, region, and tenancy, in exchange for a significant discount when compared to on-demand pricing.

Now the downside with RIs is that they generally offer pretty limited flexibility. So let's say, for instance, your requirements change and you need to use a new instance type, or a different AWS region, before your reservation term is up. That RI you originally purchased might not be usable any longer, depending on the type of RI you have.

And just as a refresher, there are two types of Reserved Instances. Standard RIs offer the lowest cost, but are also the least flexible. They allow you to modify the availability zone or instance size within an instance family, but you can't change the instance family itself. This makes Standard RIs better suited for more stable, long-term workloads where you're confident that you won't need to make any changes over that one- or three-year term. Now in the worst-case scenario, if you commit to the wrong Standard RI and you need to get out of it before your term is over, you do have the option to resell it in AWS’s Reserved Instance Marketplace.

Then there's Convertible RIs, which offer more flexibility, but also come with a lower discount compared to Standard RIs. Convertible RIs can be exchanged for another Convertible RI with different attributes, including a different instance family, size, operating system, or tenancy. The trade-off here is that your savings will be less than they would be with a Standard RI. Convertible RIs make more sense when you know your needs are going to evolve over time, but you still want some of the security and savings that come with making a long-term commitment. Convertible RIs cannot be resold in the Reserved Instance Marketplace.

AWS offers some form of Reserved Instance or Reserve Node capacity across several key services besides just EC2. These can provide significant cost savings for your predictable workloads that leverage the Amazon Relational Database Service or RDS, Amazon OpenSearch Service, ElastiCache, and Redshift. And while the specific implementation and flexibility of the Reserved Instances or Reserved Nodes across these services may vary, they all provide meaningful cost savings in exchange for a one- or three-year commitment.

So that's Reserved Instances.

Now after Reserved Instances came Savings Plans, and AWS introduced the Savings Plan model to address some of these flexibility issues with RIs that we just discussed. Savings Plans offer a more moderate approach to commitment-based pricing by allowing you to commit to spending a certain amount of dollars per hour on compute services at a discounted rate. And again, this is over a one- or three-year term.

The key difference here is that this usage isn't tied to a specific instance type. In fact, it may not even be tied to EC2 at all. Think of Savings Plans as offering commitment without as many constraints. With Savings Plans, AWS will automatically apply your committed spend amount to any eligible compute usage. And just like there are two types of RIs, it turns out that there are two types of Savings Plans as well. We have Compute Savings Plans, which apply to all EC2 instance types across regions and platforms, as well as to Fargate and Lambda for serverless compute. These Savings Plans provide the most flexibility and are a great choice if your workloads are evolving.

And there are also EC2 Instance Savings Plans, which offer greater savings but less flexibility, as they will only apply to a specific EC2 instance family within a single region. Now, these do still give you some freedom within an instance family if you need to change things like the operating system or the instance size within that family. But if you're heavily tied to EC2 and you're confident that you'll need to continue running your workloads on a specific instance family in a specific region for years to come, these will offer greater savings than Compute Savings Plans. And it's worth pointing out that AWS also offers SageMaker Savings Plans for your machine learning workloads.

But there are no Savings Plans for the other services I mentioned earlier that offer some form of Reserved Instances or reserved nodes, like RDS, OpenSearch, ElastiCache, or Redshift. If you want to maximize your savings with any of those services, you'll have to use Reserved Instances.

So that was a quick overview of cost commitment models. Now let's see how to choose which model is right for your workloads.

AWS Cost Commitment Models

About