Pod Network Benchmark Topology — same node · same AZ · cross-AZ

Pod Network Benchmark Topology — same node · same AZ · cross-AZ An architecture diagram generated by Archify. srv-same · 10.0.2.72 · same node as cli · AZ ap-northeast-2a › node ip-10-0-2-128 · m5.xlarge srv-same 10.0.2.72 · same node as cli cli · 10.0.2.109 · iperf3 · fortio · ping · AZ ap-northeast-2a › node ip-10-0-2-128 · m5.xlarge cli 10.0.2.109 · iperf3 · fortio · ping srv-a · 10.0.2.37 · same AZ · other node · AZ ap-northeast-2a › node ip-10-0-2-20 · m5.xlarge srv-a 10.0.2.37 · same AZ · other node srv-b · 10.0.3.65 · other AZ · AZ ap-northeast-2b › node ip-10-0-3-32 · m5.xlarge srv-b 10.0.3.65 · other AZ same node · RTT 0.040 ms · p50 0.259 ms same AZ · RTT 0.339 ms · p50 0.461 ms cross-AZ · RTT 0.544 ms · p50 0.704 ms AZ ap-northeast-2a node ip-10-0-2-128 · m5.xlarge node ip-10-0-2-20 · m5.xlarge AZ ap-northeast-2b node ip-10-0-3-32 · m5.xlarge Legend Client pod Server pod (iperf3 -s + fortio server)

RTT / HTTP p50 ladder

  • • ping avg 0.040 → 0.339 → 0.544 ms (same node → same AZ → cross-AZ)
  • • HTTP p50 0.259 → 0.461 → 0.704 ms (100 qps · 4 conn · keepalive)
  • • One AZ hop = RTT +0.21 ms, p50 +0.24 ms

Bandwidth caps (iperf3 TCP · 20 s)

  • • Single inter-node flow 4.96 Gbps — identical same-AZ and cross-AZ (EC2 5 Gbps per-flow cap)
  • • 8 flows 9.94 Gbps = m5.xlarge 10 Gbps peak
  • • Same node 29.97 Gbps (1 flow, client CPU 99.8 %) / 48.15 Gbps (8 flows)

Cross-AZ cost anchor

  • • 180 s × 9.93 Gbps = 223.4 GB → $2.23 out + $2.23 in = $4.47
  • • Regional Data Transfer $0.01/GB — billed in each direction across the AZ boundary
  • • No step-down toward the 1.25 Gbps baseline observed within 180 s