Performance Benchmarks¶

Hornet is built on winnow, a zero-copy, monomorphising parser combinator library designed for throughput. Benchmarks are measured with Criterion.rs (100 statistical samples per group).

Test environment¶

Property	Value
Platform	macOS 15 (Apple M-series)
Rust	stable (1.86+)
Profile	`release` (`opt-level = 3`)
Criterion warm-up	3 s
Criterion measurement	5 s

Note

Linux x86_64 builds show comparable throughput. CI benchmarks run on ubuntu-latest and results are uploaded as workflow artifacts for reference.

`named.conf` — parsing¶

Seven fixture sizes from a trivial single-block config up to a 50 000-zone stress test. The first five (tiny → xlarge) are run in standard CI; the last two are run via make bench-stress only.

Fixture	Input size	Median time	Throughput	CI?
`tiny` — single options block	60 B	596 ns	102 MiB/s	✓
`small` — options + ACL + 3 zones + logging	~700 B	7.9 µs	87 MiB/s	✓
`medium` — views, ACLs, keys, logging, 11 zones	~2.8 KB	34 µs	81 MiB/s	✓
`large` — 100 zones	~12 KB	218 µs	54 MiB/s	✓
`xlarge` — 1 000 zones	~124 KB	13.8 ms	8.7 MiB/s	✓
`xxlarge` — 10 000 zones (stress)	~1.3 MB	1.19 s	1.02 MiB/s	—
`xxxlarge` — 50 000 zones (stress)	~6.5 MB	30.5 s	207 KiB/s	—

Typical production configs

Real-world named.conf files rarely exceed 50 KB. At the medium fixture scale (a realistic multi-view production config), hornet parses in under 35 µs.

Scaling behaviour beyond 1 000 zones

Throughput drops significantly past ~1 000 zones due to heap allocation pressure and CPU cache spill as the in-memory AST grows. For configs in that range, consider splitting across multiple include files or using the view mechanism to partition zones.

`named.conf` — writing and validation¶

Using the medium fixture (~2.8 KB, 11 zones across 2 views).

Operation	Median time	Throughput
`write_named_conf`	12.9 µs	212 MiB/s
`validate_named_conf`	967 ns	2.77 GiB/s
Round-trip (parse + write)	46.6 µs	59 MiB/s

Zone files — parsing¶

Five fixture sizes from a minimal zone (SOA + NS + 2 A records) up to 10 000 host records.

Fixture	Input size	Median time	Throughput
`tiny` — SOA + NS + 2 A records	~200 B	2.1 µs	85 MiB/s
`small` — 20 records, all major types	~1 KB	8.4 µs	134 MiB/s
`medium` — 100 A records	~5.5 KB	24.7 µs	91 MiB/s
`large` — 1 000 A records	~55 KB	228 µs	99 MiB/s
`xlarge` — 10 000 A records	~550 KB	2.3 ms	104 MiB/s

Consistent linear scaling

Zone file throughput stays in the 85–106 MiB/s band across four orders of magnitude of input size, demonstrating the linear complexity of the winnow parser combinators.

Zone files — writing and validation¶

Using the medium fixture (100 records, ~5.5 KB).

Operation	Median time	Throughput
`write_zone_file`	18.5 µs	121 MiB/s
`validate_zone_file`	110 ns	19.7 GiB/s
Round-trip (parse + write)	45.4 µs	49 MiB/s

Running the benchmarks yourself¶

# Full run (uses Criterion defaults: 3 s warm-up, 5 s measurement)
make bench

# Abbreviated run, standard fixtures only (used in CI)
make bench-quick

# Stress fixtures only — 10k and 50k zones (slow)
make bench-stress

# Compile only — verify bench code builds without running
make bench-compile

Criterion generates an interactive HTML report at target/criterion/report/index.html.

Comparing against a baseline¶

# Save a baseline labelled "main"
cargo bench -- --save-baseline main

# Later, compare a branch against it
cargo bench -- --load-baseline main --baseline main

Source¶

Benchmark source lives in benches/:

benches/named_conf.rs — parse, write, validate, round-trip (tiny → xlarge 1k zones)
benches/named_conf_stress.rs — stress fixtures (10k and 50k zones, not run in CI)
benches/zone_file.rs — parse, write, validate, round-trip for zone files