Rightsize Kubernetes Memory Over-Provisioned Workload

This recommendation identifies Kubernetes workloads where memory is over-provisioned by 30% or more based on actual memory utilization patterns over the last 30 days.

📘

This recommendations type requires version 1.2 or later of the CloudZero Agent for Kubernetes

What it does

• Analyzes memory utilization vs. memory resource requests for Kubernetes pods
• Identifies workloads using less than 70% of their requested memory (30%+ over-provisioned)
• Calculates potential cost savings from rightsizing memory requests
• Uses 30-day historical utilization data with data quality filters to ensure reliable recommendations

How savings are calculated

Savings Formula: workload_cost × (% memory over-provisioned - 30%)

Examples:

• 40% memory over-provisioned workload → 10% of workload cost as savings
• 50% memory over-provisioned workload → 20% of workload cost as savings
• 60% memory over-provisioned workload → 30% of workload cost as savings

Recommended Actions

• Review memory resource requests for identified workloads
• Reduce memory requests to match actual utilization patterns (with appropriate safety margin)
• Test application performance after rightsizing to ensure requirements are met
• Monitor workload performance and memory utilization after changes
• Consider implementing Vertical Pod Autoscaler (VPA) for automatic rightsizing, or Horizontal Pod Autoscaling (HPA) to reduce replica count
• Update deployment manifests and Helm charts with optimized memory requests