Skip to main content

​
EKS Prefix Delegation - IP Planning Guide

kubectl get nodes -o custom-columns=NAME:.metadata.name,CAPACITY:.status.capacity.pods,ALLOCATABLE:.status.allocatable.pods,AGE:.metadata.creationTimestamp

​
📊 Your Subnet Configuration

SubnetCIDRAvailable IPsAvailability Zone
Subnet 110.50.64.0/204,096 IPsAZ-1
Subnet 210.50.112.0/204,096 IPsAZ-2
Subnet 310.50.16.0/204,096 IPsAZ-3
Total Available: ~12,288 IPs across all subnets

​
🔄 How Prefix Delegation Works

​
Traditional Secondary IP Mode (Before)

Node ENI → Requests individual IPs
├─ Primary IP: 1 IP (for node)
├─ Secondary IP: 1 IP (for pod 1)
├─ Secondary IP: 1 IP (for pod 2)
└─ Secondary IP: 1 IP (for pod 3)

​
Prefix Delegation Mode (Now)

Node ENI → Requests /28 prefix (16 IPs at once)
├─ Primary IP: 1 IP (for node)
├─ Prefix 1: /28 = 16 IPs (for 16 pods)
├─ Prefix 2: /28 = 16 IPs (for 16 pods)
└─ Prefix 3: /28 = 16 IPs (for 16 pods)

​
🖥️ Your Node Configuration (m6g.large)

​
Network Interface Limits:

  • ENIs per node: 3 ENIs maximum
  • Prefixes per ENI: 16 prefixes (with prefix delegation)
  • IPs per prefix: 16 IPs (/28 CIDR block)

​
IP Allocation per Node:

m6g.large Node
├─ ENI 0 (Primary)
│   ├─ 1 Primary IP (node itself)
│   └─ Up to 16 prefixes × 16 IPs = 256 IPs
│
├─ ENI 1 (Secondary)
│   └─ Up to 16 prefixes × 16 IPs = 256 IPs
│
└─ ENI 2 (Secondary)
    └─ Up to 16 prefixes × 16 IPs = 256 IPs

Total theoretical: 1 + (3 ENIs × 16 prefixes × 16 IPs) = 769 IPs
But Kubernetes limits it to ~110 pods per node (kubelet default)

​
💡 How Your Configuration Works

​
With WARM_PREFIX_TARGET = “1”

When a node starts:
  1. Initial allocation: Requests 1 prefix (/28 = 16 IPs) per ENI
    • 3 ENIs × 1 prefix × 16 IPs = 48 IPs reserved initially
    • Plus 1 primary IP for node = 49 IPs per node
  2. As pods are scheduled:
    • When 15 out of 16 IPs in a prefix are used
    • VPC CNI requests another /28 prefix (16 more IPs)
    • Keeps 1 “warm” prefix ready at all times
  3. Maximum allocation per node:
    • Up to 110 pods (Kubernetes limit)
    • Each pod = 1 IP
    • Actual IPs requested = ~7-8 prefixes × 16 = 112-128 IPs per node

​
📈 Your Capacity Calculation

​
Current Node Group (1-2 nodes):

NodesIPs Used (Initial)IPs Used (Full)Subnet Impact
2 nodes2 × 49 = 98 IPs2 × 128 = 256 IPsMinimal

​
Scaled Node Group (Max capacity scenarios):

Let’s calculate different scenarios:

​
Scenario 1: 10 Nodes

  • Initial: 10 × 49 = 490 IPs
  • Full capacity: 10 × 128 = 1,280 IPs
  • Pods: 10 × 110 = 1,100 pods

​
Scenario 2: 50 Nodes

  • Initial: 50 × 49 = 2,450 IPs
  • Full capacity: 50 × 128 = 6,400 IPs
  • Pods: 50 × 110 = 5,500 pods

​
Scenario 3: 100 Nodes

  • Initial: 100 × 49 = 4,900 IPs
  • Full capacity: 100 × 128 = 12,800 IPs
  • Pods: 100 × 110 = 11,000 pods

​
✅ Your Subnet Capacity Assessment

​
Per Subnet (each /20):

Available IPs per subnet: 4,096
Maximum nodes per subnet (at full capacity): ~32 nodes
Maximum pods per subnet: 32 × 110 = ~3,520 pods

​
Total Capacity (all 3 subnets):

Total Available IPs: 12,288
Maximum nodes (distributed): ~96 nodes
Maximum pods: 96 × 110 = ~10,560 pods

​
🎯 Your Configuration is EXCELLENT! ✅

​
Why Your Setup Works Well:

  1. ✅ Large Subnets: /20 subnets are very generous
  2. ✅ Multiple AZs: Distributes load across 3 availability zones
  3. ✅ Room to Grow: Can support ~30 nodes per subnet
  4. ✅ Prefix Efficiency: /28 prefixes align well with your subnet size

​
🔍 Real-World Example

Let’s say you have 2 nodes (current config):

​
Node 1 in Subnet 1 (10.50.64.0/20):

- Node IP: 10.50.64.10
- Prefix 1 (ENI 0): 10.50.64.16/28 → 10.50.64.16 - 10.50.64.31 (16 IPs)
- Prefix 2 (ENI 1): 10.50.64.32/28 → 10.50.64.32 - 10.50.64.47 (16 IPs)
- Prefix 3 (ENI 2): 10.50.64.48/28 → 10.50.64.48 - 10.50.64.63 (16 IPs)

​
Node 2 in Subnet 2 (10.50.112.0/20):

- Node IP: 10.50.112.10
- Prefix 1 (ENI 0): 10.50.112.16/28 → 10.50.112.16 - 10.50.112.31 (16 IPs)
- Prefix 2 (ENI 1): 10.50.112.32/28 → 10.50.112.32 - 10.50.112.47 (16 IPs)
- Prefix 3 (ENI 2): 10.50.112.48/28 → 10.50.112.48 - 10.50.112.63 (16 IPs)
Each pod gets one IP from these prefixes automatically!

​
📋 Recommendations

​
✅ You’re Good to Go!

Your current setup can comfortably support:
  • Short term: 2-10 nodes (your current plan)
  • Medium term: 20-30 nodes per subnet
  • Long term: Up to ~90 nodes total

​
Optional Optimizations:

If you want to be even more efficient, you could adjust WARM_PREFIX_TARGET: 
# More aggressive (saves IPs)
WARM_PREFIX_TARGET = "0"  # Request prefixes only when needed

# Current (balanced)
WARM_PREFIX_TARGET = "1"  # Keep 1 warm prefix ready

# More prefixes (faster pod startup)
WARM_PREFIX_TARGET = "2"  # Keep 2 warm prefixes ready

​
🚀 Summary

You have PLENTY of IP space! Your /20 subnets are well-sized for prefix delegation. With your current 2-node setup, you’re using less than 0.5% of your subnet capacity. Even scaling to 50+ nodes would still be comfortable.

​
📚 Additional Resources