Skip to main content

Infrastructure Performance Benchmark

3 min read
AI documentation guide

This is a measurement plan for EKS networking, DNS and autoscaling. The items below define test scope; this document does not present measured results.

Network Performance​

Cilium ENI vs VPC CNI Comparison​

A separate document examines historical summaries for five VPC CNI and Cilium configurations and explains their verification limits.

See 1. CNI Performance Comparison for metric definitions and requirements for a new experiment. Raw logs and effective configuration are missing, so a product performance ranking remains unverified.

Configurations requested by the historical runner (5):

  • A: VPC CNI Default (kube-proxy + iptables)
  • B: Cilium + kube-proxy (Overlay)
  • C: Cilium kube-proxy-less (eBPF replacement)
  • D: Cilium ENI Mode (Native Routing)
  • E: Cilium ENI with additional options (DSR, native XDP, socket LB, etc.; effective activation unverified)

Gateway API Performance​

Metrics

  • Request processing latency (P50, P95, P99)
  • Requests per second (RPS)
  • TLS handshake overhead

DNS Performance​

CoreDNS Optimization Before/After Comparison​

Metrics

  • DNS resolution latency
  • Cache hit rate
  • Queries per second throughput
  • NodeLocal DNSCache effect

Autoscaling Response Speed​

Karpenter vs Cluster Autoscaler​

Metrics

  • Scale-out response time
  • Node provisioning time
  • Pod scheduling delay

Cost Efficiency​

Cost-to-Performance Ratio by Instance Type​

Metrics

  • Throughput per vCPU
  • Cost per memory
  • Spot vs On-Demand stability comparison