← Back to comparison

Numa DNS

Port: 15454 | Image: numa:bench

Results by Scenario

Cache hit

All queries resolve from server cache — measures peak throughput.

Provider QPS Avg (ms) Lost % Total Qrys Base Mem (MB)
Quad9 951.66 98.09 0.13 28679 16.50
Cloudflare 1004.34 92.58 0.13 30262 18.65
Google 958.93 95.06 0.19 28913 16.61
OpenDNS 996.54 93.88 0.20 30033 14.95
Quad9 916.65 96.96 0.26 27652 15.08
Cloudflare 977.00 95.17 0.14 29441 15.70
Google 906.56 96.36 0.31 27360 15.80
OpenDNS 960.24 97.17 0.14 28937 15.91

Cache miss

Every query requires full upstream resolution — measures worst-case performance. In common-denominator mode some servers report near-zero QPS when plain-DNS upstream forwarding fails.

Provider QPS Avg (ms) Lost % Total Qrys Base Mem (MB)
Quad9 197.47 112.45 0.26 5981 17.67
Cloudflare 199.45 102.62 0.11 6000 19.63
Google 197.85 102.91 0.23 5960 17.57
OpenDNS 199.38 108.17 0.13 6000 16.78
Quad9 199.35 110.78 0.13 6000 15.15
Cloudflare 199.55 100.57 0.03 6000 16.50
Google 199.51 99.94 0.06 6000 16.63
OpenDNS 199.47 105.49 0.09 6000 16.86

Mixed

90% cached + 10% unique domains — simulates real-world mixed workload.

Provider QPS Avg (ms) Lost % Total Qrys Base Mem (MB)
Quad9 946.41 95.95 0.20 28533 16.97
Cloudflare 954.24 94.05 0.22 28773 18.53
Google 971.45 94.59 0.19 29289 16.66
OpenDNS 909.77 97.41 0.25 27444 16.92
Quad9 951.02 98.67 0.12 28659 16.08
Cloudflare 913.28 98.28 0.23 27547 15.58
Google 953.22 96.13 0.19 28735 16.01
OpenDNS 993.95 94.19 0.14 29956 16.55

Blocked

Domains on the sinkhole blocklist — measures filtering throughput.

Provider QPS Avg (ms) Lost % Total Qrys Base Mem (MB)
Quad9 919.16 98.80 0.20 27718 17.01
Cloudflare 930.97 96.00 0.23 28075 18.48
Google 915.50 95.27 0.29 27624 16.49
OpenDNS 959.27 92.71 0.26 28937 16.72
Quad9 932.12 97.70 0.19 28099 15.59
Cloudflare 924.91 94.29 0.31 27915 15.88
Google 934.38 93.43 0.27 28192 15.99
OpenDNS 938.13 97.79 0.18 28281 15.89

Methodology

Transport Protocol Confound

This server uses DoT for best-of-breed mode. Comparisons against servers using different protocols (DoH, DoT, plain) partly reflect protocol overhead. The common-denominator mode forces all servers to plain UDP 53 to isolate server-side performance.

Warmup

Before each scenario, a 5-second dnsperf warmup pass is run to stabilize connections and fill the TCP congestion window. For cache-hit and mixed scenarios, the server cache is also pre-warmed. The measurement window (-l 30) begins only after both warmup phases complete.

Test Parameters