Right size Windows workloads

Overview

Right sizing is one of the most potent cost-saving tools. AWS offers various methods to collect right sizing information, from reviewing potential workloads by using an AWS Optimization and Licensing Assessment (AWS OLA) to reviewing existing workloads by using AWS Cost Explorer.

This section shows you how to use AWS Compute Optimizer to identify Amazon EC2 right sizing opportunities. Compute Optimizer helps prevent over-provisioning and under-provisioning for the following types of AWS resources:

Cost optimization scenario

Measuring the effectiveness of right sizing can be challenging, because right sizing efforts can be directed toward a specific app, team, or the entire organization. For example, consider an organization that migrates several thousand instances to AWS, with 90 percent of their fleet comprising Windows workloads. The organization can employ Compute Optimizer to analyze their fleet and discover significant over-provisioning across their accounts and AWS Regions. Then, they can use AWS Systems Manager Automation to right size their fleet through multiple maintenance windows. As a result, the organization manages to adjust the right sized instance type for 70 percent of their fleet and achieves a cost savings of 35 percent.

The following dashboard illustrates the savings achieved over several months as this example organization strategically implemented the right sizing recommendations of Compute Optimizer. Their objective was to operate their existing workloads as efficiently as possible in order to resume a stalled migration from a colocation data center nearing the end of its contract.

Cost optimization recommendations

We recommend that you take the following next steps to optimize your costs by using Compute Optimizer:

Enable Compute Optimizer

You can enable Compute Optimizer at the organization or single account level in AWS Organizations. The organization-wide configuration provides ongoing reports for new and existing instances across your entire fleet for all member accounts. This enables right sizing to be a recurring activity instead of a point-in-time activity.

Organization level

For most organizations, the most efficient way to use Compute Optimizer is at the organization level. This provides multi-account and multi-Region visibility into your organization and centralizes the data into one source for review. To enable this at the organization level, do the following:

Single account level

If you're targeting an account with high costs but don't have access to AWS Organizations, you can still enable Compute Optimizer for that account and Region. To learn about the opt-in process, see Getting started with AWS Compute Optimizer in the Compute Optimizer documentation.

Enable memory metrics collection for Windows nodes

Memory metrics supply Compute Optimizer with the essential metrics required to make well-informed right sizing recommendations in your organization. This is due to the analysis of CPU, memory, network, and storage being conducted before offering a recommendation.

To pass memory metrics from Windows EC2 instances to Compute Optimizer, you must enable the CloudWatch agent and configure memory metrics to be collected every 60 seconds. There is no additional cost for using memory metrics with CloudWatch.

Enable the CloudWatch agent and configure memory metrics

Download the ComputeOptimize.yml file. You can use this file to enable memory collection for all instances in your account. The template file generates the following components:

Next, do the following:

Monitor the progress of the association

View the metrics in CloudWatch

We recommend that you wait for at least five minutes for the metrics to populate CloudWatch.

Consume Compute Optimizer recommendations

Consider an example that focuses on making right sizing changes within a single account and single Region. In this example, Compute Optimizer is enabled at the organization level across all accounts. Keep in mind that right sizing is a disruptive process that in most cases is carried out with precision by the application owners during a scheduled maintenance window over several weeks.

If you navigate to Compute Optimizer from within an organization's management account (as shown in the following steps), you can choose the account that you want to investigate. In this example, there are six instances running in a single account in the us-east-1 Region. All six instances are over-provisioned. The goal is to resize the instances based on the recommendations from Compute Optimizer.

Identify over-provisioned instances and export recommendation details

Choose instances based on recommendations

Instance recommendations are based on the performance metrics collected and analyzed by Compute Optimizer. It's essential to be aware of the workloads running on the instance to ensure that you choose the best instance. This example assumes that you can choose from the latest generation of Amazon EC2 R6i, R5, and T3 instances. T3 instances are burstable and have lower network bandwidth capabilities. R5 and R6 instances have the same cost per hour and are nearly identical. However, the R6 instance has a higher network bandwidth capacity, features the latest generation of Intel processors, and offers the same compute footprint as the R5. In this example, R6 is the best option to choose for resizing.

Tag instances for right sizing

Tagging your workloads is a powerful tool for organizing your resources in AWS. Tags enable you to gain fine-grained visibility into costs and facilitate chargeback. For more information about strategies and methods for adding tags to AWS resources, see the AWS Whitepaper Best Practices for Tagging AWS Resources. For this example, you can use the AWS Tag Editor to make tagging adjustments across the over-provisioned instances that you wish to target for resizing during a maintenance window. You can also use this tag to view the costs before and after the change.

After creating and applying user-defined tags to your resources, it may take up to 24 hours for the tags to appear on your cost allocation tags page for activation. After you select your tags for activation, it could take another 24 hours for the tags to become active.

For advanced users, you can use AWS CloudShell within the target account and Region to tag multiple instances. For example:

bash
#!/bin/bash
# Set variables
TAG_KEY="rightsizing"
TAG_VALUE="type-m5"
# Get a list of instance IDs
INSTANCE_IDS=$(aws ec2 describe-instances —query "Reservations[].Instances[].InstanceId" —output text)
# Loop through each instance ID and add the tag
for INSTANCE_ID in $INSTANCE_IDS; do
 aws ec2 create-tags —resources $INSTANCE_ID —tags Key=$TAG_KEY,Value=$TAG_VALUE
done

Enable the cost allocation tag to work with AWS billing tools

We recommend activating the user-defined cost allocation tag. This enables the Rightsizing tag to be recognized and filterable in the AWS billing tools (for example, Cost Explorer and AWS Cost and Usage Report). If you don't enable this, the tag filtering option and data won't be available. For information about using cost allocation tags, see Activating user-defined cost allocation tags in the AWS Billing and Cost Management documentation.

After 24 hours, the tag should appear in Cost Explorer.

Implement right sizing recommendations with Systems Manager Automation

Resizing is a scenario that requires an instance to be stopped and started. In this scenario, you might need to handle this interruption in a maintenance window and need different teams to handle their own resizing. Before you change an instance type, review the Considerations for compatible instance types in the Amazon EC2 documentation.

The example steps in this section implement right sizing recommendations per account and Region by using a Systems Manager Automation document called AWS-ResizeInstance. This approach is typical for most organizations as most organizations require different instance types for different purposes. You can also use the same AWS-ResizeInstance automation document to target single and multi-account deployments.

Consider alternative resizing methods

If you're using a launch template to deploy your instances, you can update the launch template with the right sized instance type, and then perform an instance refresh to replace the instances with the right-sized version.

If you plan to use the right sizing process across multiple accounts and Regions, you must create a custom Systems Manager Automation document. This document enables you to feed in multiple instances as a parameter and target instances moving to the same destination instance type (for example, all instances transitioning to t3a.medium, irrespective of the source instance type).

Review before and after costs in Cost Explorer

After you have right sized your resources, you can use Cost Explorer to show before and after costs by using the Rightsizing tag. Recall that you can use resource tags to track costs. By using several layers of tags, you can achieve granular visibility into your costs. In the example covered in this guide, the Rightsizing tag is used to apply a generic tag to all targeted instances. Then, a team tag is used to further organize resources. The next step is to introduce application tags to further show the cost impact for operating a specific application.

The following diagram shows the tag structure for an organization.

Consider the example of a business that right sizes the production web servers owned by the Operations team. In Cost Explorer, the Rightsizing tag is set to enabled, and the Team tag is set to operations. In this example, the right sizing effort reduces operating costs from 0.89 cents to 0.28 cents an hour. Assuming 744 hours per month, the annual cost before right sizing is $7,945.92. After right sizing, the annual cost drops to $2,499.84. This translates to a 68.5 percent decrease in annual workload costs. Imagine the impact of this across a large organization. Keep in mind, this is done in a sample environment and the instances are mostly idle. In a production environment, you can see savings between 10–35 percent.

Now, consider the impact of right sizing the production bastion host owned by the Engineering team. In Cost Explorer, the Rightsizing tag is set to enabled, and the Team tag is set to engineering. In this example, the right sizing effort reduces costs from 0.75 cents to 0.44 cents an hour. Assuming 744 hours per month, the annual cost before right sizing is $6,696.00. After right sizing, the annual cost drops to $3,928.32.

If you use multiple tags, you can filter the data down to granular cost details. In this example, the Team tag reduces the noise so that you can view the impact at a team level. Because the Rightsizing tag is enabled, you can also filter for any instance that has that tag with the value of enabled or no value present. This can provide a global view of your right sizing efforts, particularly when viewed in the management account (payer) at the Cost Explorer level. This view enables you to see all accounts and instances.

Consider an example at the single account level where the Rightsizing tag is set to enabled.  The operating costs drop from $1.64 an hour to $.72 cents an hour. Assuming 744 hours per month, the annual cost before right sizing is $14,641.92. After right sizing, the annual cost drops to $6,428.16. This translates to a 56 percent decrease in compute costs for this account.

Before embarking on your right sizing journey, consider the following:

Recommendations

We recommend the following next steps:

Additional resources