The Fundamentals of AWS Pricing
Amazon Web Services (AWS) is a top cloud provider, offering many services, including storage, artificial intelligence (AI) and machine learning (ML), networking, and more. Each service is priced differently, and there are many pricing models available.
The most flexible of the six AWS pricing models is the “pay as you go” model, which lets you pay only for the resources you use and quickly scale as needed. There is no need for long-term contracts. The basic option lets you pay for what you actually use. Other pricing models enable you to pay less when reserving resources in advance or get volume discounts when you use more resources.
6 AWS Pricing Models
Understanding AWS pricing models is critical for optimizing your Amazon costs.
AWS Free Tier
AWS offers a free tier that lets you sample a wide range of products for free. Here are the three types of free offerings:
- Always free—this option lets you use a wide range of AWS resources for free. There is a limitation on the amount of resources you can use, and as long as you do not exceed it you can continue using the service without a time limitation. For example, you can get one million AWS Lambda requests per month.
- 12 months free—the count starts after the initial sign-up date to AWS. This model provides you with free resources for each service for a limited time period. For example, you can get 5 GB of standard storage in Amazon Simple Storage Service (Amazon S3).
- Short-term free trial—AWS provides a free trial, which you can use to try out certain services. For example, you can use Amazon Redshift for free for a duration of two months.
Note that additional features not included in the free tier may incur additional charges. You can learn more in the official free tier page.
The on-demand pricing model lets you pay by the second or by the hour without any upfront payment. There is no need to commit in advance. You can start using the service and pay as needed. However, while on-demand pricing is usually the most flexible model it is also often the most expensive option. It is ideal for unpredictable workloads but not for long term use.
AWS offers major discounts on their available and unused compute capacity. This model is called Spot Instances. You can get discounts of up to 90% for using spot instances—this is the largest possible discount available on AWS.
To use spot instances, you need to set up a bid, and when an instance is available at that price you can get a discount on using spare capacity. You can also leverage the Hibernate option, which keeps your instance in the same state and launches it again when there is spare capacity.
While Spot Instances offers major discounts, it also has its limitations. The most concerning limitation is that once the instance is no longer available, it is terminated. You get a 2-minutes notice and then your instance is dropped. This is not ideal for demanding workloads. You can set up AWS monitoring to receive “rebalancing warnings” which are sometimes sent by Amazon in advance of the two minute notice period.
Reserved Instances (RIs) provide discounts in return for an upfront commitment to use these resources for a long duration of 1-3 years. This model lets you purchase instances in advance, and use them as needed. AWS provides several payment options, including No Upfront, Partial Upfront, and All Upfront. The more you pay in advance, the bigger the discount.
There are two main types of RIs—Standard RI and Convertible RI. Each comes with different options and several instance families. Convertible RIs let you commit only to a specific region for a certain period of time and change the instance family as needed. This level of flexibility makes Convertible RIs more expensive than the Standard RIs.
RIs are initially less difficult to use than Spot Instances and offer great discounts. However, as you accumulate more RIs, management becomes complex. Additionally, it can be difficult to choose the right instance family and you must pay for the capacity you committed to even if you later do not need it.
Saving Plans let you get a high discount rate in return for committing to use AWS resources for a duration of one to three years. You pay an hourly rate, and the discount is subtracted from the on-demand pricing of your resources.
AWS offers three main payment options for Saving Plans—Upfront, Partial Upfront, and All Upfront. You can choose one option or a combination of the three and leverage the discounts across your workloads.
Saving Plans are applied in the account they are purchased from, and then can be applied across all connected accounts. This ensures you can get discounts for several different services, rather than just one. If you want more control over how discounts are applied, you can use an empty account to purchase all available Saving Plans.
A dedicated host is a physical server offered for rent. AWS lets you rent an entire server, which is reserved for your use only. You gain access to the server and AWS is responsible for administrative tasks, including hardware maintenance and cleaning.
Dedicated Hosts offer a secure and reliable option for running workloads on AWS. However, this service can be highly expensive, which is why it is used primarily by enterprises.
In this article I explained the fundamentals of pricing in Amazon’s cloud services, including the following pricing models:
- Free tier – letting you experience some Amazon services free forever, and others free for a limited time and quantity
- On-demand – paying per hour or second of usage for Amazon resources
- Spot instances – deeply discounted instances from Amazon’s spare capacity, which can be interrupted on short notice
- Reserved instances – letting you commit to computing capacity for a period of 1 or 3 years and receive substantial discounts
- Savings plans – committing to a certain level of spend on Amazon across your organization in exchange for a discount
- Dedicated hosts – a physical server provided for your use within the Amazon data center
I hope this will be of help as you plan your use of Amazon and other cloud services.