Performance and Benchmarks

This page collects the performance notes moved out of the README and keeps public benchmark snapshots by version. The goal is not to claim an absolute winner, but to show the performance profile of ForgeDNS under different levels of policy complexity, concurrency, and transport-path pressure.

What ForgeDNS Cares About

ForgeDNS is not only interested in peak numbers for the simplest possible case. The more relevant questions are:

• Is the hot path still controlled when cache, rules, fallback, and rewrites are enabled?
• Is overall latency acceptable when several upstreams race concurrently?
• Can the structure still be optimized after adding more protocols and plugins?
• Do system integrations and observability stay off the critical response path enough to avoid dragging performance down?

Metric Notes

• Higher QPS is better
• Lower average latency and latency stddev are better
• run_dnsperf_compare.sh is better for saturated or higher-concurrency throughput and queueing behavior
• run_dnsperf_latency_compare.sh is better for low-concurrency latency, with clients == outstanding
• These snapshots are better read as version-specific distributions of strengths against mosdns
• Because the 2026-04-13 compare pack updated the scenario catalog, query sets, and some workload definitions, the absolute numbers in v0.1.0 and v0.3.0 should not be treated as a direct regression chart across versions

Legend:

• Green means ForgeDNS performs better on that metric
• Red means mosdns performs better on that metric
• Neutral means the gap is small and shown only as a reading aid, not as a claim of statistical significance

v0.3.0

Higher-Concurrency Throughput and Average Latency

Test environment:

• Date: 2026-04-13
• System: Linux 6.8.12-2-pve x86_64
• Selector: core
• Compared versions: ForgeDNS v0.3.0, mosdns v5.3.4-0-gb732318

Load-test parameters:

• Tool: dnsperf
• warmup_seconds=2
• bench_seconds=8
• bench_repeats=3
• dnsperf_clients=32
• dnsperf_threads=4
• dnsperf_outstanding=1024
• dnsperf_max_qps=unlimited

How to read these results:

• Scenarios such as baseline UDP forward, concurrent upstreams, and dual-entry UDP/TCP include upstream forwarding or upstream races, so they mainly reflect end-to-end proxy behavior; upstream RTT, upstream response stability, and race strategy matter more here than small local processing differences
• cache hotpath, local answers, and server local UDP/TCP are closer to local processing cost
• domain set and composite provider chain are better indicators for rule-heavy and dataset-heavy policy workloads

The table below shows the per-scenario medians aggregated from repeats:

Scenario	ForgeDNS QPS	mosdns QPS	QPS Delta	ForgeDNS Avg Latency	mosdns Avg Latency
baseline UDP forward	35,498.1	36,883.2	-3.8%	7.735 ms	11.244 ms
cache hotpath	139,133.9	134,881.6	+3.1%	0.637 ms	0.721 ms
dual-entry UDP	35,800.9	35,382.3	+1.2%	7.092 ms	10.170 ms
dual-entry TCP	37,295.1	37,221.2	+0.2%	24.646 ms	25.083 ms
concurrent upstreams	21,038.7	13,319.4	+58.0%	10.404 ms	20.601 ms
local answers	126,268.4	149,119.6	-15.3%	0.783 ms	0.639 ms
domain set	165,647.7	36,383.7	+355.3%	0.549 ms	4.078 ms
ip set	133,355.1	150,756.7	-11.5%	0.740 ms	0.637 ms
composite provider chain	158,693.6	25,972.4	+511.0%	0.532 ms	6.251 ms

Low-Concurrency Latency Sweep

Test environment:

• Date: 2026-04-13
• System: Linux 6.8.12-2-pve x86_64
• Selector: latency-core
• Compared versions: ForgeDNS v0.3.0, mosdns v5.3.4-0-gb732318

Load-test parameters:

• Tool: dnsperf
• warmup_seconds=1
• bench_seconds=5
• bench_repeats=3
• latency_client_levels=1 2 4
• dnsperf_threads=1
• dnsperf_timeout=5
• dnsperf_outstanding=matches_client_count

The three tables below focus on average latency and jitter first, while QPS remains a secondary sanity metric. They use the same color rule as above: lower latency or jitter is green, higher latency or jitter is red, and ties are neutral.

clients=1, outstanding=1

Scenario	ForgeDNS Avg Latency	mosdns Avg Latency	Latency Delta	ForgeDNS Jitter	mosdns Jitter	Jitter Delta
baseline UDP forward	6.716 ms	5.708 ms	+17.66%	0.170 ms	1.140 ms	-85.09%
cache hotpath	0.029 ms	0.031 ms	-6.45%	0.017 ms	0.021 ms	-19.05%
dual-entry UDP	5.949 ms	6.394 ms	-6.96%	0.426 ms	2.364 ms	-81.98%
dual-entry TCP	5.984 ms	6.118 ms	-2.19%	0.380 ms	0.829 ms	-54.16%
local answers	0.026 ms	0.029 ms	-10.34%	0.016 ms	0.019 ms	-15.79%
domain set	0.025 ms	0.108 ms	-76.85%	0.011 ms	0.095 ms	-88.42%
composite provider chain	0.025 ms	0.156 ms	-83.97%	0.012 ms	0.129 ms	-90.70%
server local UDP	0.025 ms	0.027 ms	-7.41%	0.013 ms	0.013 ms	+0.00%
server local TCP	0.027 ms	0.030 ms	-10.00%	0.009 ms	0.018 ms	-50.00%

clients=2, outstanding=2

Scenario	ForgeDNS Avg Latency	mosdns Avg Latency	Latency Delta	ForgeDNS Jitter	mosdns Jitter	Jitter Delta
baseline UDP forward	6.001 ms	7.382 ms	-18.71%	0.262 ms	0.858 ms	-69.46%
cache hotpath	0.033 ms	0.037 ms	-10.81%	0.050 ms	0.053 ms	-5.66%
dual-entry UDP	6.408 ms	5.923 ms	+8.19%	1.247 ms	0.488 ms	+155.53%
dual-entry TCP	5.651 ms	5.633 ms	+0.32%	0.417 ms	0.518 ms	-19.50%
local answers	0.040 ms	0.031 ms	+29.03%	0.021 ms	0.018 ms	+16.67%
domain set	0.029 ms	0.111 ms	-73.87%	0.013 ms	0.104 ms	-87.50%
composite provider chain	0.029 ms	0.165 ms	-82.42%	0.014 ms	0.141 ms	-90.07%
server local UDP	0.030 ms	0.029 ms	+3.45%	0.016 ms	0.016 ms	+0.00%
server local TCP	0.028 ms	0.030 ms	-6.67%	0.014 ms	0.015 ms	-6.67%

clients=4, outstanding=4

Scenario	ForgeDNS Avg Latency	mosdns Avg Latency	Latency Delta	ForgeDNS Jitter	mosdns Jitter	Jitter Delta
baseline UDP forward	5.977 ms	5.910 ms	+1.13%	0.355 ms	2.700 ms	-86.85%
cache hotpath	0.044 ms	0.060 ms	-26.67%	0.028 ms	0.063 ms	-55.56%
dual-entry UDP	6.637 ms	5.426 ms	+22.32%	25.556 ms	0.435 ms	+5774.94%
dual-entry TCP	6.451 ms	6.941 ms	-7.06%	4.422 ms	26.437 ms	-83.27%
local answers	0.056 ms	0.040 ms	+40.00%	0.030 ms	0.025 ms	+20.00%
domain set	0.034 ms	0.155 ms	-78.06%	0.016 ms	0.141 ms	-88.65%
composite provider chain	0.034 ms	0.221 ms	-84.62%	0.015 ms	0.172 ms	-91.28%
server local UDP	0.042 ms	0.038 ms	+10.53%	0.024 ms	0.024 ms	+0.00%
server local TCP	0.042 ms	0.039 ms	+7.69%	0.026 ms	0.022 ms	+18.18%

v0.3.0 Result Readout

• If your real deployment relies on complex rule sets, large datasets, or multi-upstream racing, ForgeDNS is in a stronger position in v0.3.0; domain set, composite provider chain, and concurrent upstreams are all clearly ahead.
• cache hotpath now shows a small lead for ForgeDNS, which means the gap on high-frequency cache-hit traffic has narrowed substantially.
• In local-answer workloads, the gap between ForgeDNS and mosdns is already fairly small; the ip set result is also more constrained by the local-answer performance ceiling than by a completely separate bottleneck.
• For forward, dual-entry, and concurrent upstreams, the numbers should be read as end-to-end proxy behavior. Upstream latency and upstream response stability are the dominant factors there, not just local implementation overhead.
• In the low-concurrency latency sweep, domain set, composite provider chain, cache hotpath, and server local TCP are the most stable strong areas; dual-entry UDP shows noticeably higher jitter at clients=4, so that scenario still needs more stability work.

v0.1.0

The following public result set is preserved as the historical v0.1.0 snapshot from March 26, 2026.

Higher-Concurrency Throughput and Average Latency

Test environment:

• CPU: Intel N100, 4 cores
• Memory: 1 GB
• Environment: LXC inside a PVE VM
• System: Linux 6.8.12-2-pve x86_64
• Date: 2026-03-26
• Compared versions: forgedns v0.1.0, mosdns v5.3.4-0-gb732318

Load-test parameters:

• Tool: dnsperf
• warmup_seconds=2
• bench_seconds=8
• bench_repeats=3
• dnsperf_clients=32
• dnsperf_threads=4
• dnsperf_outstanding=1024
• dnsperf_max_qps=unlimited

The table below shows the average of three runs for each scenario:

Scenario	ForgeDNS QPS	mosdns QPS	QPS Delta	ForgeDNS Avg Latency	mosdns Avg Latency
baseline UDP forward	37,789.6	37,269.2	+1.4%	9.142 ms	12.312 ms
cache hotpath	131,982.3	133,380.3	-1.0%	1.235 ms	0.696 ms
dual-entry UDP	39,614.4	34,356.8	+15.3%	8.946 ms	10.009 ms
dual-entry TCP	36,257.9	35,975.4	+0.8%	25.403 ms	25.577 ms
concurrent upstreams	21,694.8	13,195.4	+64.4%	15.065 ms	23.790 ms
fallback standby	22,259.9	23,223.9	-4.2%	16.376 ms	10.616 ms
local answers	132,286.6	146,754.3	-9.9%	1.250 ms	0.636 ms
DoH upstream (HTTP/2)	29,781.6	25,835.7	+15.3%	13.363 ms	11.445 ms
domain set	172,061.7	35,966.1	+378.4%	0.901 ms	4.210 ms
ip set	134,257.4	150,923.0	-11.0%	1.227 ms	0.625 ms
sequence base	131,995.6	150,301.5	-12.2%	1.265 ms	0.622 ms
match true	135,326.0	153,289.5	-11.7%	1.217 ms	0.629 ms
match false	136,740.1	152,297.5	-10.2%	1.201 ms	0.630 ms
match qname	132,289.4	152,203.6	-13.1%	1.248 ms	0.638 ms

v0.1.0 Historical Readout

• ForgeDNS was stronger in concurrent upstream races, DoH upstreams, dual-entry UDP, and large domain_set scenarios
• mosdns was still faster in cache-hit paths, local answers, basic sequence, and lighter matcher scenarios
• fallback standby still showed room for further optimization

Raw Materials

• Benchmark directory: benchmarks/mosdns_compare/README.md
• Scenario list: benchmarks/mosdns_compare/scenarios.tsv
• Higher-concurrency script: benchmarks/mosdns_compare/run_dnsperf_compare.sh
• Low-concurrency latency script: benchmarks/mosdns_compare/run_dnsperf_latency_compare.sh