Performance — node-http-server

Recorded medians

Targeted work per second and retained memory.

These medians come from one recorded environment. Each row keeps its nine raw baseline and candidate samples in the linked JSON.

9Alternating samples

Node 24.18.0Runtime

Windows x64Platform

Core Ultra 9 275HXProcessor

Workload	9.0.2 median	9.1.0 median	Result	Primary change
1,000 repeated query values	1,076 req/s	7,695 req/s	7.15× speedup	In-place array accumulation.
1,000-domain virtual-host miss	13,926 req/s	23,259 req/s	1.67× speedup	Compiled O(1) hostname map.
1,000-domain virtual-host hit	4,173 req/s	4,198 req/s	1.01× speedup	Hostname map plus canonical-root reuse.
16-byte custom-hook range from 8 MiB	790 req/s	2,310 req/s	2.92× speedup	Exact selected-byte read.
Default Brotli for 2.5 MiB structured text	0.580 req/s	112.267 req/s	193.56× speedup	Validated default quality 4; 161,802 → 270,759 compressed bytes.
100,000 cold Config constructions	201,737 instances/s	580,412 instances/s	2.88× speedup	Deferred MIME-map materialization.
10,000 cold Configs retained	11,816.5 B/instance	5,786.0 B/instance	51.0% lower	Deferred per-instance MIME map.
Immediate deploy + close	180 leaked listeners	0 leaked listeners	180 → 0	Pending listeners close cleanly.

Ratio meaning: each displayed ratio is the magnitude of change; its speedup, slowdown, reduction, or increase label states the direction. The listener row reports the correctness count across 180 recorded cycles per source.

Compatibility and representative paths

Read the optimized paths beside their controls.

Workload	9.0.2 median	9.1.0 median	Result	Contract
Default-hook static GET	4,055 req/s	4,254 req/s	1.05× speedup	Exact static response through the normal route.
Bodyless static HEAD	10,981 req/s	13,739 req/s	1.25× speedup	GET metadata with an empty wire body.
Bodyless custom request hook	16,698 req/s	16,507 req/s	1.01× slowdown	Empty string and Buffer body contract.
Default-hook HEAD for 8 MiB	5,254 req/s	5,884 req/s	1.12× speedup	Full metadata and an empty wire body through the bodyless path.
Custom beforeServe HEAD	765 req/s	688 req/s	1.11× slowdown	The hook receives the full 8 MiB representation; the wire body stays empty.
Brotli quality 11	0.584 req/s	0.581 req/s	1.01× slowdown	`server.brotliQuality:11` preserves the high-quality option through Node's built-in `node:zlib`; runtime dependencies stay at zero.
Cold one-key MIME overlays retained	11,818.6 B/instance	5,972.4 B/instance	49.5% lower	Overlays stay deferred and instance-isolated.
Materialized MIME maps retained	11,816.7 B/instance	12,018.8 B/instance	1.7% higher	Reading `contentType` creates the ordinary isolated map.

The two 8 MiB HEAD rows quantify the full-representation compatibility cost for custom hooks. The materialized Config row quantifies the point where deferred state becomes the complete per-instance MIME map.

Workload contracts

Every timed path proves its output.

Area	Standard profile	Acceptance contract
HTTP	400 small requests, 40 repeated-query requests, 8 large-file requests, or 2 Brotli requests per sample; 10 clients.	Status, headers, body bytes, decompressed bytes, and response counts.
Query	1,000 ordered values assigned to one key.	Array length, first value, and last value.
Routing	1,000 mixed-case domains with an accepted Host and an unknown Host.	Selected root returns exact bytes; miss returns `421`.
Range hook	16 bytes selected from the middle of an 8 MiB file.	Hook sees 16 bytes, transforms them, preserves `206`/`Content-Range`, and runs `afterServe` once.
Config	100,000 constructions or 10,000 retained instances in fresh garbage-collected workers.	Default checksum, key order, own MIME entries, mutation isolation, and false-to-overlay restoration.
Lifecycle	20 deploy-on-port-zero plus immediate-close cycles per sample.	Live listener and ready-callback counts after close.

Comparison method

Keep order, processes, and evidence explicit.

Method	Implementation
Exact baseline	Exports `Server.js`, `Config.js`, and `MimeTypes.js` from Git tag `9.0.2` into an OS temporary directory.
Alternating order	Each sample alternates baseline-first and candidate-first execution.
Process isolation	HTTP variants use separate server processes. Config memory and lifecycle use fresh `--expose-gc` workers.
Recorded boundary	Fixture creation, startup, response preflight, warmup, worker startup, and cleanup occur outside action timing.
Aggregation	Median, minimum, maximum, and all nine raw samples are retained.
Source identity	Git ref, commit, working-tree state, source hashes, and benchmark-harness hashes identify the complete comparison.
Environment	Node, V8, platform, OS release, architecture, processor, logical cores, and total memory are recorded.

Interpretation: compare ratios within the recorded workload and environment. Run the same profile on deployment-class hardware when capacity planning needs host-specific numbers.

Reproduce

Run the exact source comparison.

shell · repository checkout with tag 9.0.2

npm run benchmark:core

node --expose-gc benchmark/core.js \
    --output site/benchmarks/core-9.0.2-vs-9.1.0.json

File	Role
`benchmark/core.js`	Exports the baseline, creates fixtures, alternates samples, validates results, and writes JSON.
`benchmark/core-server.js`	Runs one source variant and exposes the hook scenarios through IPC.
`benchmark/core-worker.js`	Measures Config construction/memory and immediate-close correctness in fresh processes.
`site/benchmarks/core-9.0.2-vs-9.1.0.json`	Published environment, configuration, summaries, source hashes, and raw samples.

Open raw JSON →Read every benchmark option →Run the operational benchmark →Testing & coverage →