EKS Network Performance Tuning Order

EKS Network Performance Tuning Order A workflow diagram generated by Archify. 01 / Infrastructure & kernel optimization 02 / Placement & policy optimization Pick instance type · C5n · M5n · R5n · Infrastructure & kernel optimization Pick instance type C5n · M5n · R5n Choose CNI mode · VPC CNI · custom · Infrastructure & kernel optimization Choose CNI mode VPC CNI · custom Tune MTU · Jumbo frames 9001 · Infrastructure & kernel optimization Tune MTU Jumbo frames 9001 Tune TCP · Scaling · BBR · Infrastructure & kernel optimization Tune TCP Scaling · BBR Placement locality · Affinity · topology · Placement & policy optimization Placement locality Affinity · topology Trim net policies · Fewer, narrower · Placement & policy optimization Trim net policies Fewer, narrower bandwidth set CNI fixed kernel tuned final pass Legend Node & kernel tuning Policy AWS resource

Instances and CNI

  • • Pick types like C5n, M5n, or R5n that offer 25 Gbps or more plus enhanced networking (SR-IOV)
  • • Use WARM_IP_TARGET and MINIMUM_IP_TARGET to cut IP allocation latency
  • • If IPs run short, attach a secondary CIDR with custom networking

MTU and TCP

  • • Jumbo frames (9001 MTU) cut packet overhead, but the MTU must match along the whole path
  • • Enable TCP window scaling and consider switching congestion control to BBR
  • • Match TCP buffer sizes and keepalive values to the workload

Placement and monitoring

  • • Keep chatty pods in one AZ with pod affinity and topology spread constraints
  • • Topology-aware routing removes cross-AZ hops
  • • Compare before and after with VPC Flow Logs and CloudWatch metrics, then verify with iperf3