Testing mode.
Abstract network background
Beta Benchmarks

BreathGSLB Performance

Real-world performance benchmarks demonstrating production-class throughput with sub-millisecond latency.

Executive Summary

BreathGSLB delivers exceptional performance in real-world conditions, achieving 70k+ QPS with sub-millisecond latency and zero packet loss.

Peak Performance

70k+
Queries Per Second

Ultra-low Latency

<1ms
Average Response Time

Rock-solid Reliability

99.99%
Uptime Guarantee

Benchmark Results

Comprehensive performance testing conducted over real network paths with DNSSEC enabled.

Performance Summary

ScenarioQueries/sAvg LatencyLoss
60k QPS59,999.970.29 ms0%
70k QPS69,999.950.38 ms0%
~67k QPS (limit)66,887.260.59 ms0.18%
174k QPS stress174,236.8015.22 ms9.61%
140k QPS stress139,716.0716.31 ms18.89%

QPS vs Avg Latency

QPS vs Loss %

Technical Analysis

Understanding the methodology and significance of these results.

What's Being Measured

Early dnsperf runs against BreathGSLB as an authoritative nameserver with DNSSEC enabled (ECDSA-P256) and EDNS buffer 1232. Tests use IPv6 first. The goal is to show stable, repeatable throughput on real hardware and a real network path—not lab-only loopback numbers.

Test Setup (Summarized)

  • Remote path (headline numbers): 10 Gb worker → 10 Gb router → server ("homer") on a 2.5 Gb NIC.
  • Cloud baseline: AWS t2.micro/t2.nano authoritative.
  • Local loopback/LAN: Used only to sanity-check software limits, not published as headline throughput.
  • Software/build: BreathGSLB 0.0.1.ff81290 (Linux), DNSSEC ECDSA-P256, NSEC (NSEC3 later), dnsperf 2.14.0.
  • Query mix: ~60% NOERROR / ~40% NXDOMAIN (small answers typical of authority service).
  • Run length: 30 s per sweep.
  • Logging: Query logging disabled during perf runs.

Why These Numbers Matter

  • Production-class performance: ~70k QPS at ~0% loss over a real network path with DNSSEC on is production-class for an authoritative server.
  • Headroom for intelligence: Sub-millisecond average latency at 60–70k QPS shows there's still headroom for failover logic and GeoDNS policy.
  • Graceful behavior under stress: When pushed past the envelope, the path drops packets as expected for high-PPS UDP; the daemon stays healthy.

Understanding Stress Test Results

Those 140–174k QPS "stress" rows deliberately over-drove the harness:

  • -q (outstanding queries) and thread counts were set far beyond what the RTT and NIC queues can sustain.
  • UDP has no backpressure. Oversized bursts overflow kernel/NIC queues, so some queries time out even if the server is fine.
  • A clean rule of thumb: Size the pipeline to QPS × RTT × 2–4. With ~1.2 ms RTT, 70k QPS only needs a few thousand outstanding queries in total. Right-sized runs (60k/70k) show 0% loss.

Reproducing Clean Runs

dnsperf -s ns2-gslb.akadata.ltd -d queries.txt -l 30 \
  -Q 60000 -q 2000 -T 8 -c 8 -S 5

dnsperf -s ns2-gslb.akadata.ltd -d queries.txt -l 30 \
  -Q 70000 -q 2500 -T 8 -c 8 -S 5

Tips: Disable query logging; keep EDNS buf at 1232; ensure UDP socket and NIC buffers are sized sensibly; if chasing beyond ~70k over a single NIC, use two load clients or multiple SO_REUSEPORT workers.

Key Insights

What these benchmarks reveal about BreathGSLB's capabilities.

Production Readiness

Achieving 70k+ QPS with sub-millisecond latency over real network paths demonstrates that BreathGSLB is ready for production deployment in demanding environments.

Intelligent Headroom

Sub-millisecond latency at peak throughput indicates ample capacity for advanced GSLB features like health checks, failover logic, and GeoDNS policies without performance degradation.

Predictable Behavior

Graceful degradation under stress with predictable packet loss patterns validates the robustness of both the software and network infrastructure design.

Scalable Architecture

Linear performance scaling and efficient resource utilization demonstrate an architecture designed for horizontal scaling to meet any enterprise DNS demands.

Experience DNS & GSLB Freedom

Take control of your DNS infrastructure with BreathGSLB. Run it yourself for complete autonomy, or let Breath Technology host it for you. The choice is yours—and so is the freedom.

Self-Hosted

Deploy on your own infrastructure with complete control and customization.

  • • Full administrative control
  • • Zero vendor lock-in
  • • Custom policy implementation
  • • On-premises deployment

Hosted Solution

support.

  • • 24/7 expert support
  • • Automatic updates & maintenance
  • • Global anycast network