Rightsizing Recommendations That Name a Target Size

Concrete or nothing

A rightsizing recommendation in ZopNight states the current size and the target. If the engine cannot name a target, or the target would be the same as the current size, it does not produce a resize card at all. That rule removes the vague “consider a smaller instance” advice that nobody acts on.

The evidence is measured utilization over a real window, pulled from CloudWatch, Cloud Monitoring or Azure Monitor, and the saving is priced from the difference between the current and target rates.

Examples from the catalog

Compute

• EC2 Rightsizing, Graviton Migration Opportunity, EC2 Older Instance Family
• GCP VM Oversized - Downsize to Smaller Predefined Machine Type
• Azure VM Rightsizing, Downscale Opportunity
• GPU Instance Low Utilization

Databases

• RDS Instance Right-sizing, GCP Cloud SQL Right-sizing, Azure SQL Right-sizing
• RDS Dev/Test Burstable Opportunity, Azure MySQL Flexible Server Burstable Tier Opportunity

Serverless and containers

• Lambda Over-Provisioned Memory, ECS Fargate Oversized Task, Cloud Run Service Low Concurrency
• GKE Node Pool VM-Size Rightsizing, AKS Node Pool Underutilized

Storage

• EBS gp2 to gp3 Migration, EBS Provisioned IOPS Over-Allocation, Azure Premium Disk IOPS Downgrade, GCP Disk SSD with Low IOPS

Applying a resize

Several of these can be applied from ZopNight. A hot resize (an EBS volume’s IOPS, an Azure disk tier, a SageMaker endpoint) runs as a remediation step. A resize that needs the resource stopped first, such as an EC2 or Compute Engine type change, runs as a stop, resize and start, and is never automatic on a stateful resource like a managed database.

After you apply one, ZopNight compares the resource’s shape against the target it recommended, which was frozen when the finding was raised. A resize that was reverted, or that landed on a different size, is visible as such.

Your own thresholds

Built-in rules not strict enough, or too strict? A watch policy lets you define your own signal, such as average CPU below a threshold over a number of days for a scope of resources, with a rightsize outcome. ZopNight then recommends a concrete one-tier-down target with the real price difference.

See rightsizing recommendations by cloud, rightsizing methodology, and how resizes are applied.