Right-Sizing Instances

CPU Utilization (CPUUtilization)

• Primary metric for compute right-sizing
• Average < 40% over extended period suggests over-provisioning
• Consistent > 80% may indicate under-provisioning
• Analyze peaks, not just averages

Memory Utilization (Custom Metric)

• Requires CloudWatch Agent installation
• Critical for memory-intensive workloads
• Default EC2 metrics do NOT include memory
• Install unified CloudWatch agent for collection

Network I/O (NetworkIn, NetworkOut)

• Important for network-bound applications
• Helps identify if network bandwidth is a bottleneck
• Consider enhanced networking instance types

Disk I/O (DiskReadOps, DiskWriteOps, EBSReadOps, EBSWriteOps)

• Identifies storage performance requirements
• May indicate need for different EBS volume types
• Consider instance store for high IOPS workloads

Disk Queue Depth (DiskReadBytes, DiskWriteBytes)

• High queue depth indicates storage bottleneck
• May need larger instance with more EBS bandwidth

Why Install CloudWatch Agent

• Memory utilization not available by default
• Disk space utilization requires agent
• Process-level metrics available
• Custom application metrics

Key Memory Metrics Collected:

• mem_used_percent
• mem_available
• mem_cached
• swap_used_percent

Configuration Example:

{
  "metrics": {
    "metrics_collected": {
      "mem": {
        "measurement": ["mem_used_percent"]
      },
      "disk": {
        "measurement": ["used_percent"],
        "resources": ["/"]
      }
    }
  }
}

Best Practice: Deploy CloudWatch agent to all EC2 instances for comprehensive right-sizing data

What It Does

• Uses machine learning to analyze resource utilization
• Provides recommendations for EC2, Auto Scaling groups, EBS, and Lambda
• Compares current configuration against 140+ instance types
• Projects future utilization based on historical patterns

Supported Resources:

• EC2 instances
• EC2 Auto Scaling groups
• Amazon EBS volumes
• AWS Lambda functions
• Amazon ECS services on Fargate

Recommendation Types: | Finding | Description | |---------|-------------| | Under-provisioned | Current instance too small for workload | | Over-provisioned | Current instance larger than needed | | Optimized | Current instance appropriately sized | | Not applicable | Insufficient data or unsupported |

Data Requirements:

• Minimum 30 consecutive hours of utilization data
• Recommended: 14 days for accurate recommendations
• More historical data = better recommendations

Standard vs Enhanced Infrastructure Metrics

Standard (Free):

• Uses CloudWatch default metrics
• 5-minute data granularity
• Basic recommendation accuracy
• Memory utilization not included

Enhanced (Paid):

• Uses CloudWatch agent metrics
• 1-minute data granularity
• Includes memory utilization
• Higher recommendation accuracy
• External metrics support
• Up to 3 months lookback period

Enabling Enhanced Recommendations:

• Requires CloudWatch Agent deployment
• Activate in Compute Optimizer settings
• Additional CloudWatch charges apply
• Recommended for production workloads

Savings Estimation:

• Provides estimated monthly savings
• Shows risk level for each recommendation
• Compares multiple instance options

What It Provides

• Cost-focused right-sizing recommendations
• Based on CloudWatch utilization metrics
• Integrated with AWS billing data
• Shows actual cost savings potential

Key Features:

• Identifies instances with low utilization
• Recommends specific instance type changes
• Calculates monthly savings estimate
• Links recommendations to pricing models

Recommendation Criteria:

• CPU utilization below threshold (configurable)
• Network utilization patterns
• EBS bandwidth requirements
• Historical usage analysis (14 days minimum)

Accessing Recommendations:

1. AWS Console > Cost Explorer
2. Right Sizing Recommendations (left menu)
3. Filter by linked account, tags, or instance type
4. Sort by estimated savings

Step 1: Identify Candidates

• Review Compute Optimizer recommendations
• Check Cost Explorer right-sizing suggestions
• Analyze CloudWatch metrics for low utilization
• Focus on largest potential savings first

Step 2: Validate Requirements

• Verify application compatibility with new instance type
• Check for architecture changes (x86 vs ARM/Graviton)
• Confirm EBS bandwidth requirements
• Validate network performance needs

Step 3: Test Changes

• Test in non-production environment first
• Run performance benchmarks
• Validate application functionality
• Compare before/after metrics

Step 4: Implement Change

• Schedule maintenance window
• Stop instance (for EBS-backed)
• Change instance type
• Start instance and validate

Step 5: Monitor and Iterate

• Monitor new utilization levels
• Adjust if issues arise
• Document changes for auditing
• Schedule next review cycle

Compatibility Factors:

Virtualization Type:

• Some older AMIs require paravirtual (PV)
• Modern instances require HVM
• Check AMI virtualization type before changing

EBS Optimization:

• Some instance types are EBS-optimized by default
• Others require explicit enablement
• Verify EBS bandwidth meets application needs

Enhanced Networking:

• Requires ENA or Intel 82599 VF support
• May require driver updates
• Check instance type compatibility

Instance Store:

• Not all instance types have instance store
• Instance store data lost on stop
• Ensure storage strategy accounts for this

Processor Architecture:

• x86_64 (Intel/AMD) vs arm64 (Graviton)
• Graviton offers up to 40% better price/performance
• Requires application compatibility testing

Establish Regular Review Cadence

• Monthly: Review Compute Optimizer recommendations
• Quarterly: Comprehensive right-sizing analysis
• Annually: Architecture review with right-sizing

Automate Where Possible

• Set up CloudWatch alarms for underutilization
• Use AWS Lambda for automated reporting
• Integrate recommendations into ticketing systems
• Schedule automated Compute Optimizer exports

Create Right-Sizing Governance

• Define utilization targets (e.g., 70% CPU average)
• Establish approval workflow for changes
• Track savings realized over time
• Include right-sizing in cloud governance policy

Integrate with DevOps

• Include right-sizing in deployment pipelines
• Validate instance sizing during code reviews
• Use Infrastructure as Code for consistency
• Test performance during CI/CD

Special Considerations for ASG

Analyze Aggregate Metrics:

• Look at group-level utilization, not individual instances
• Consider scaling patterns and frequency
• Review scaling policy effectiveness

Mixed Instance Policies:

• Use multiple instance types for flexibility
• Let ASG optimize placement
• Reduce over-provisioning for peak capacity

Optimization Strategies:

• Right-size minimum capacity first
• Adjust scaling thresholds based on utilization
• Consider predictive scaling for known patterns

Implementation Approach:

1. Update launch template with new instance type
2. Set desired capacity to force replacement
3. Or use instance refresh for gradual rollout
4. Monitor during transition