Performance lab / Run 024
NTS deployment performance
Explore how query load and NTS key establishment compete for CPU across bare metal, containers, and Kubernetes.
Best throughput
72K qps
Kubernetes
Lowest P99
0.98 ms
end-to-end model
Peak CPU
78.3%
within envelope
Failed / minute
0
no modeled loss
Deployment envelope
Current point estimate · one-minute observation window
mode: model
| Deployment | Throughput | P50 / P95 / P99 | CPU | Stage load | Bottleneck |
|---|---|---|---|---|---|
| Bare Metal | 72,000 qps | 0.4 / 0.68 / 0.98 ms | 61.8% | Query 59.3% · KE 2.6% | Headroom available |
| Docker | 72,000 qps | 0.45 / 0.79 / 1.14 ms | 68.5% | Query 65.7% · KE 2.9% | Headroom available |
| Kubernetes | 72,000 qps | 0.54 / 0.99 / 1.51 ms | 78.3% | Query 75% · KE 3.3% | Headroom available |
P99 latency under load
Queueing knee appears as capacity is exceeded
Throughput delivered
Requested load versus processed time queries
Cost decomposition
Key establishment vs. time queries
TLS overhead affects the key-establishment stage only. Increasing active clients raises periodic handshake demand; increasing query rate raises the cookie-authenticated NTP stage.
KE: 27 TLS/sQueries: 72K NTP/s