EKS Pod Density Tuning

Default pod-per-node limits are conservative. The tuning that doubles density without breaking the network.

The default limit

EKS pod density is the number of pods that can run on a single EC2 node. The default is constrained by AWS's VPC CNI, which assigns one IP per pod from the node's ENI capacity. The math produces low pod density on small instances; tuning the CNI dramatically improves density and reduces node count.

What the default limit looks like:

• EKS uses ENI-based networking.: The AWS VPC CNI assigns each pod a private IP from the node's VPC subnet. Pods get real VPC IPs; the network is flat. The simplicity is valuable; the cost is per-pod IP consumption.
• Pods per node formula.: Pods per node equals (ENIs per instance times IPs per ENI) minus 1. The minus 1 reserves one IP for the node itself. The formula produces specific limits per instance type.
• Small instances: 8 to 15 pods.: A t3.medium has 3 ENIs with 6 IPs each: 17 pods minus 1 equals 16 pods, but the actual EKS default is 17 due to documentation differences. Many small instance types are limited to 8 to 15 pods. The density is low.
• Limits density.: Low pod density means more nodes for the same workload. More nodes mean more cost. The default limits become a real cost driver at scale; teams pay for nodes they did not need.
• Per instance type lookup.: The exact pods-per-node limit is documented per instance type. Teams selecting instances should consider not just compute and memory but also pod density.

The default limit is the foundation. Most teams hit it before they hit CPU or memory limits on their nodes.

Prefix delegation

Prefix delegation is the AWS VPC CNI feature that solves the pod density problem. Each ENI claims a /28 IP range (16 IPs) instead of a single IP. The math changes dramatically: instead of pods-per-node being limited by ENI count times IPs per ENI, it becomes ENI count times 16 times IPs per ENI.

• Each ENI can claim /28 IP ranges.: A /28 prefix is 16 contiguous IPs. The ENI claims the prefix; pods on that ENI get IPs from the prefix. Single-IP allocation is replaced by prefix allocation.
• Multiplies pod count 16x.: The pod density on the same node increases roughly 16-fold. A node that supported 16 pods now supports 250 plus pods (subject to other limits like CPU and memory). The density is no longer ENI-limited.
• Enable in CNI config.: The configuration is a single setting in the AWS VPC CNI. ENABLE_PREFIX_DELEGATION=true. The setting takes effect on new pods; existing pods continue with the old allocation until rescheduled.
• Test with worker pool first.: The change is significant. Test on a non-production worker pool first; verify pods schedule correctly; verify networking works. Production rollout follows the validation.
• VPC CIDR planning.: Prefix delegation consumes VPC IP space faster (16 IPs per ENI claim vs 1 per pod). VPCs with tight IP allocation may run out of space; planning is required.

Prefix delegation is the standard solution. Most production EKS clusters now use it; the cost savings are too significant to ignore.

Savings

The savings from pod density tuning are large and immediate. Fewer nodes for the same workload means lower compute cost; the savings compound across the fleet.

• Higher pod density equals fewer nodes.: The math is simple. If density doubles, nodes halve. The compute cost scales directly with node count; the savings are direct.
• Typically 30 to 50% node count reduction.: Real-world deployments report 30 to 50 percent node count reduction after enabling prefix delegation. The exact savings depend on the workload profile (small pods benefit most; large pods see less benefit).
• Savings on associated costs too.: Fewer nodes mean lower NAT egress costs (fewer node IPs), lower per-node management overhead, lower licensing costs for per-node tools. The savings extend beyond direct compute.
• Operational simplification.: A smaller fleet is simpler to operate. Fewer nodes to monitor, fewer to upgrade, fewer to debug when something goes wrong. The operational savings are real but harder to quantify.
• Continuous benefit.: The savings continue every month. The one-time cost of enabling prefix delegation produces ongoing savings; the ROI is immediate and continuous.

EKS pod density tuning is one of those Kubernetes operations disciplines that pays off proportionally to the cluster size. Nova AI Ops integrates with EKS cluster telemetry, surfaces pod density patterns, and produces the optimization queue that the platform team uses to drive node count down without affecting workloads.