Day 9: DynamoDB Capacity Planning

Phase 5 · Aug 20, 2026

Agenda (2–3 hours)

• Read (45 min): DynamoDB capacity documentation; DynamoDB pricing page; AWS "Choosing the Right DynamoDB Capacity Mode" blog post
• Study (45 min): Calculate RCU/WCU for a table with 1KB average item size at 500 reads/s and 100 writes/s; compare on-demand vs provisioned cost
• Practice (45 min): Enable CloudWatch contributor insights on a DynamoDB table; identify the top 5 partition keys by access volume
• Challenge (30 min): Design a capacity auto-scaling policy for a table with highly variable traffic (1-100× daily variation)

Read and Write Capacity Units

WCU (Write Capacity Unit):

• 1 WCU = 1 write/s of up to 1KB
• Item > 1KB: round up to nearest KB
• Example: 500 writes/s × 2KB items = 1,000 WCU

RCU (Read Capacity Unit):

• 1 RCU = 1 strongly-consistent read/s of up to 4KB
• Eventually consistent reads consume ½ RCU
• Example: 500 reads/s × 1KB items = 125 RCU (eventually consistent)

Burst capacity: DynamoDB retains unused capacity for short spikes (up to 5 minutes worth of unused capacity)

Capacity Modes

Provisioned:

• You specify RCU/WCU targets
• Throttled when you exceed provisioned capacity
• Auto-scaling: DynamoDB adjusts within defined min/max range
• Cost: ~$0.00065/RCU/hr + $0.00013/WCU/hr

On-Demand:

• No capacity planning; AWS scales automatically
• Pay per request: ~$1.25 per million writes, $0.25 per million reads
• Good for unpredictable workloads; 2-10× more expensive at sustained load

Rule of thumb: provisioned + auto-scaling for predictable production workloads; on-demand for dev/test or highly variable new services.

Auto-Scaling Configuration

# CloudFormation
TableAutoScalingRole:
  Type: AWS::IAM::Role
  # ... policy allowing application-autoscaling to update DynamoDB capacity

WriteScalingPolicy:
  Type: AWS::ApplicationAutoScaling::ScalingPolicy
  Properties:
    PolicyType: TargetTrackingScaling
    TargetTrackingScalingPolicyConfiguration:
      TargetValue: 70  # target 70% utilization
      PredefinedMetricSpecification:
        PredefinedMetricType: DynamoDBWriteCapacityUtilization

Target 70% utilization: leaves headroom for traffic spikes; scaling lags ~2 minutes.

Adaptive Capacity

DynamoDB Adaptive Capacity automatically shifts capacity from cold partitions to hot ones:

• Works within your provisioned capacity; doesn't allow exceeding the table limit
• Handles "time-of-day" hot partition patterns automatically
• Not a substitute for good partition key design — handles imbalance, not sustained hot partitions

Key Takeaways

• RCU/WCU are the billing and quota units; calculate based on item size × request rate
• Eventually consistent reads cost half; use them when strong consistency isn't needed
• Provisioned + auto-scaling: best for sustained predictable workloads; on-demand for variable
• Adaptive capacity handles temporary hot partitions; Contributor Insights identifies persistent hot keys

Tomorrow: DynamoDB consistency and transaction patterns.