Skip to content
Ariel is an AWS Lambda designed to collect, analyze, and make recommendations about Reserved Instances for EC2.
Python TSQL Shell
Branch: master
Clone or download
meckstmd Merge pull request #7 from meckstmd/fix-issue-003-rhel-ris
Added mapping so RHEL reserved instances work.
Latest commit df19fde Aug 19, 2019
Type Name Latest commit message Commit time
Failed to load latest commit information.
ariel Add mapping so RHEL reserved instances work. Aug 12, 2019
cloudformation v2.0.0 Open Source Rollout (#8) Jun 28, 2019
schema v2.0.0 Open Source Rollout (#8) Jun 28, 2019 v2.0.0 Open Source Rollout (#8) Jun 28, 2019 v2.0.0 Open Source Rollout (#8) Jun 28, 2019
LICENSE v2.0.0 Open Source Rollout (#8) Jun 28, 2019 v2.0.0 Open Source Rollout (#8) Jun 28, 2019 Fix issue with metal instances breaking Ariel (#4) Jul 30, 2019 v2.0.0 Open Source Rollout (#8) Jun 28, 2019
config-example.yaml v2.0.0 Open Source Rollout (#8) Jun 28, 2019
requirements.txt v2.0.0 Open Source Rollout (#8) Jun 28, 2019


Ariel is an AWS Lambda designed to collect, analyze, and make recommendations about Reserved Instances for EC2.

Table of Contents


Ariel is a tool designed to make purchase recommendations for Reserved Instances for EC2. There are many tools that currently provide similar functionality to generate recommendations for either all accounts, where there is no indication of which account will benefit from the Reserved Instance, or for each account individually, which does not properly take into consideration Reserved Instance float, resulting in a recommendation of too many Reserved Instances.

The main benefit of Ariel, and what all other tools appear to lack, is the ability to combine these two approaches. Reserved Instances demand is calculated based on overall company usage, but specific recommendations are made by account such that an individual account will never purchase more than it is currently using, and also the combine Reserved Instances total will never exceed the current demand for the whole company.

Ariel also has more configuration options to allow finer tuning of the purchase recommendation algorithm; additional details can be found below in the Algorithms section.


Please refer to for installation instructions.


Configuration inputs are documented in config-example.yaml. It is expected that you will:

  • Copy and modify this file for your specific environment
  • Deploy this configuration file to S3
  • Configure the Lambda input parameter to point to this file in S3


There are many components and dependencies for Ariel that allow different deployments.

Master Billing Account - Reserved Instances

Ariel needs access to the Master Billing Account to collect all active Reserved Instances. This is collected through the Reserved Instances Utilization report provided by Cost Explorer. In addition to this report including Reserved Instances information about all active accounts in your Organization, this is one of the only ways to collect information about Reserved Instances in suspended accounts in your Organization. While suspended accounts are otherwise inactive, their Reserved Instances are still active with their monthly fee being charged to the otherwise closed account, and they are still able to float to any account linked to your Master Billing Account.

Master Billing Account - Account Names

Ariel can be configured to gather account names from the Organizations API only accessible from the Master Billing Account. This access is not required, as a flat file may be used instead, or if neither are specified, Accound IDs will be used instead of names.

Master Billing Account - Cost and Usage Reports + Athena

This version of Ariel requires Athena Optimized Cost and Usage Reports. Cost and Usage Reports can only be written to a bucket in the Master Billing Account. Athena Optimization only works when the data is stored in the bucket it was originally written to. Report data is written from a 3rd party AWS account and can not be shared with other accounts. The end result of these three statements is that Athena must be used from the Master Billing Account.

Access requirements for the Master Billing Account have been separated into a standalone CloudFormation template so that Ariel can run in an account other than the Master Billing Account, and just use a cross-account role when necessary.


Ariel does not need a VPC to run. However, if you want to deploy it to a VPC you will want to create your own VPC with the necessary Subnets, Route Tables, Internet Gateway, NAT Gateways, Network ACLs, and Security Groups, we do not provide a VPC template.

Optional Functionality

  • VPC - Ariel can be deployed with or without a VPC. Unfortunately the AWS::Serverless transform does not yet support conditionals, so the Ariel VPC and non-VPC templates are separate files.

  • AuroraDB/RDS - Ariel can be configured to publish the results to a Postgres Database. If using this database, it is recommended that you deploy Ariel in a VPC and prevent internet access directly to your database.

Recommendations Algorithm

Proper Reserved Instances management is an art form that is very difficult and complex, and mistakes can be costly. When making purchases, you really must be certain that you will use the Reserved Instances for at least as long as the break-even term for each Reserved Instance. The following content describes how to fine tune your Ariel recommendations.

Utilization vs Coverage

Reserved Instance utilization and coverage are inversely related metrics. As you increase your coverage, your utilization will usually decrease. Properly managing Reserved Instances involves maximizing coverage, while keeping utilization high enough to be cost beneficial. Additionally, since the most cost effective Reserved Instance has a three year term, you will also need to predict your future usage to make effective purchases.

For Reserved Instances purchase recommendations in Ariel, a target utilization value is specified. To achieve higher coverage, specify a lower desired utilization. Ariel also has a special BREAK_EVEN keyword for utilization that will calculate the specific break even utilization percents for the currently evaluated region and family.

Break Even

Actual break even numbers vary by region and instance type family, but for a rough rule to follow for Standard Reserved Instances in US regions:

  • A one year Reserved Instance breaks even in about 7 months.
  • A three year Reserved Instance breaks even in about 13 months.
  • Renewing a one year Reserved Instance once will cost more than if a three year Reserve Instance was purchased initially.

Aggressive / Conservative

Ariel supports two thresholds that control 3 configurations for purchasing aggressiveness. For both thresholds, the instances run rate, for an individual instance type family, region, tenancy, and operating system, is fit to a straight line, and the threshold is applied to the slope of that line.

  • If the slope is greater than AGGRESSIVE_THRESHOLD, the AGGRESSIVE configuration is used.
  • If the slope is smaller than CONSERVATIVE_THRESHOLD, the CONSERVATIVE configuration is used.
  • Otherwise, the DEFAULT configuration is used.

These thresholds are specified in "normalized units" per day. For example, a value of 1000 means an average increase of 125xlarge instances per day. Normalized units can be found in the Reserved Instances Documentation

Standard / Convertible

Convertible Reserved Instances allow for greater flexibility, but also cost on average 25% more than standard Reserved Instances. Ariel tries to give you the ability to purchase a baseline of standard Reserved Instances, while simultaneously pushing up your coverage with convertible Reserved Instances, reducing your risk if some workloads migrate away.

Convertible Reserved Instances recommendations will always be calculated after standard, and will assume that the standard Reserved Instances are also being purchased.

Slush Account

There are some cases where the total Reserved Instances recommendation by account will be less than the overall company demand. As a trivial example, if two accounts would benefit from a single Reserved Instance during different halves of the day, neither account would have sufficient demand to make the purchase independently, but the company would still benefit from purchasing one.

In this case, Ariel can be configured to make a purchase recommendation in a slush account.

The algorithm evaluates in this order:

  • Standard in-account
  • Standard slush account
  • Convertible in-account
  • Convertible slush account

Regional / Availability Zone

With the introduction of OnDemand Capacity Reservations, there is no longer any benefit from ever using Availability Zone based Reserved Instances. This support has been removed from Ariel, and Regional Reserved Instances will always be recommended.


Please refer to for detailed information about the reports generated.


Ariel is typically run daily. It's possible to configure it to run more frequently, but this provides little additional value, as Cost and Usage Reports are only updated two to three times per day. There are no concerns with running Ariel less frequently.

Some of the report data is useful to look at every day, such as the ri-usage report to find available excess Reserved Instances for shifting workloads that are easy to move across instance types. The main report from Ariel, ri-purchases, tends to have a slower cadence. It is recommended to evaluate a single ri-purchases report monthly or quarterly to make decisions about which recommendations to purchase. This review and decision making process is necessarily manual.

Assuming that a purchase is completed through a support case or your Concierge, it will take some time for your order to be processed. Ariel will automatically pick up new Reserved Instances once they have been purchased through the report accessed from Cost Explorer.

Additional Considerations

Ariel does not yet make any recommendations related to modifying existing convertible Reserved Instances. Before making a purchase you should:

  • Check the ri-usage report to find instance type families with excess unused Reserved Instances
  • Check the reserved instances report to find matching convertible Reserved Instances
  • Modify the unused Reserved Instances to cover new recommended Reserved Instances
  • Subtract the modified Reserved Instances from the recommendation


Please refer to for information about how to get involved. We welcome issues, questions, and pull requests. Pull Requests are welcome.


Sean Bastille: Micah Meckstroth:


This project is licensed under the terms of the Apache License, Version 2.0 open source license. Please refer to LICENSE for the full terms.

You can’t perform that action at this time.