Performance Test Results

Project Information

• plain-nuxt: ./test/fixtures/plain-nuxt
• i18n-micro: ./test/fixtures/i18n-micro
• i18n: ./test/fixtures/i18n
• CLI: pnpm test:performance

Description

Compares plain Nuxt, i18n-micro, and @nuxtjs/i18n under one shared dictionary profile via untestutils/perf.

Focus: build time, peak RSS, deployable code vs translations (asset) vs total, and load (Artillery).

Methodology notes

• Metrics are means of 3 consecutive runs per fixture (not interleaved).
• Artifact: Nitro node-server .output/server/index.mjs after forced nuxi build; nuxt-i18n-micro is the built module dist (not dev:prepare jiti stub). See performance-methodology.md.
• Translations = bundle.asset (locale JSON, chunks/raw/, matching locale chunks via isTranslationFile).
• plain-nuxt serves the same leaf volume as static JSON — I/O-heavy under load, not “i18n overhead”.
• Load profile --load full: warm 6s@6 + main 60s@60 uncapped (historical).
• Cool-downs --cool fast (see methodology doc).
• Artillery Error rate in tables is HTTP-status errors from the load metrics. Charts also plot vusers.failed (client timeouts under saturation); those can be high while HTTP error rate stays 0%.
• Each of the runs builds is forced (no warm-cache zeroing of build time).

Runs

All metrics below are means across 3 runs.

---

Fixture profile

Knob	Value
Locales	4 (en, de, ru, fr)
Pages	2 (index, page)
Index tree	depth 5, branch 7 → 16,807 leaf keys / locale
Secondary pages	depth 5, branch 6 → 7,776 leaf keys / page / locale
Goal	Default CLI profile (--locales 4 --keys 10000); raise knobs for regression-radar loads

Dictionaries come from test/fixtures/perf-shared/runtime.json (written by the CLI) so all three fixtures stay aligned.

Dependency Versions

Dependency	Version
node	v22.23.2
nuxt	4.5.2
nuxt-i18n-micro	3.29.8
@nuxtjs/i18n	10.6.0

Source dictionaries (pre-build)

Fixture	On-disk locale data
plain-nuxt	114.88 MB
i18n-v10	6.8 MB
i18n-micro	6.8 MB

From runtime.json (4 locales, 16,807 index leaf keys).

Build Performance for test/fixtures/plain-nuxt

• Build Time: 4.59 seconds
• Bundle Size: 8.07 MB (code: 1.26 MB, translations: 6.8 MB)
• Output dirs: public: 7 MB, server: 1.06 MB
• Max / Avg CPU: 226.73% / 155.16%
• Max / Avg Memory: 890.16 MB / 473.88 MB

Build Performance for test/fixtures/i18n

• Build Time: 5.48 seconds
• Bundle Size: 9.15 MB (code: 1.95 MB, translations: 7.21 MB)
• Output dirs: public: 323.47 KB, server: 8.84 MB
• Max / Avg CPU: 239.30% / 167.34%
• Max / Avg Memory: 1325.84 MB / 670.48 MB

Build Performance for test/fixtures/i18n-micro

• Build Time: 4.99 seconds
• Bundle Size: 8.28 MB (code: 1.48 MB, translations: 6.8 MB)
• Output dirs: public: 7.08 MB, server: 1.21 MB
• Max / Avg CPU: 232.93% / 161.92%
• Max / Avg Memory: 995.58 MB / 559.91 MB

Build Performance Summary (mean of 3 runs)

Project	Build Time	Code Bundle	Translations	Total
plain-nuxt (baseline)	4.59s	1.26 MB	6.8 MB	8.07 MB
i18n-v10	5.48s	1.95 MB	7.21 MB	9.15 MB
i18n-micro	4.99s	1.48 MB	6.8 MB	8.28 MB

“Total” = code + translations (bundle.asset). Translations include locales/, _locales/, chunks/raw/, and matching locale chunks.

Loading chart...

Loading chart...

Loading chart...

Loading chart...

Load Results for plain-nuxt

Resource Usage

• Max / Avg CPU: 142.87% / 119.18%
• Max / Avg Memory: 559.38 MB / 472.97 MB

Artillery

• Duration: 86.55s · RPS: 108.00 · Error rate: 0.00%
• Latency avg / p50 / p95 / p99: 1877.63 / 663.60 / 12547.07 / 12983.90 ms

Loading chart...

Loading chart...

Load Results for i18n-v10

Resource Usage

• Max / Avg CPU: 135.00% / 106.04%
• Max / Avg Memory: 527.66 MB / 418.26 MB

Artillery

• Duration: 78.75s · RPS: 161.33 · Error rate: 0.00%
• Latency avg / p50 / p95 / p99: 857.43 / 106.00 / 6702.60 / 6977.03 ms

Loading chart...

Loading chart...

Load Results for i18n-micro

Resource Usage

• Max / Avg CPU: 131.37% / 101.79%
• Max / Avg Memory: 474.45 MB / 305.38 MB

Artillery

• Duration: 73.18s · RPS: 251.33 · Error rate: 0.00%
• Latency avg / p50 / p95 / p99: 486.20 / 86.20 / 3534.10 / 3830.03 ms

Loading chart...

Loading chart...

Load Summary (mean of 3 runs)

Artillery

Project	Avg Response	P50	P95	P99	RPS	Error Rate
plain-nuxt	1877.63 ms	663.60 ms	12547.07 ms	12983.90 ms	108.00	0.00%
i18n-v10	857.43 ms	106.00 ms	6702.60 ms	6977.03 ms	161.33	0.00%
i18n-micro	486.20 ms	86.20 ms	3534.10 ms	3830.03 ms	251.33	0.00%

Performance Comparison

Throughput (Requests per Second)

Winner: i18n-micro with 251 RPS

Loading chart...

Latency Distribution

Winner: i18n-micro with 486.20 ms avg latency

Loading chart...

Quick Comparison

Metric	plain-nuxt	i18n-v10	i18n-micro	Best
RPS (Artillery)	108	161	251	i18n-micro
Avg Latency	1877.63 ms	857.43 ms	486.20 ms	i18n-micro
P99 Latency	12983.90 ms	6977.03 ms	3830.03 ms	i18n-micro
Error rate	0.00%	0.00%	0.00%	-

Comparison: plain-nuxt (baseline) vs i18n v10

Metric	plain-nuxt (baseline)	i18n v10	Difference
Max Memory	559.38 MB	527.66 MB	-31.72 MB
Avg Memory	472.97 MB	418.26 MB	-54.71 MB
Response Avg	1877.63 ms	857.43 ms	-1020.20 ms
Response P95	12547.07 ms	6702.60 ms	-5844.47 ms
RPS (Artillery)	108.00	161.33	+53.33
Error rate	0.00%	0.00%	0.00 %

Comparison: plain-nuxt (baseline) vs i18n-micro

Metric	plain-nuxt (baseline)	i18n-micro	Difference
Max Memory	559.38 MB	474.45 MB	-84.93 MB
Avg Memory	472.97 MB	305.38 MB	-167.59 MB
Response Avg	1877.63 ms	486.20 ms	-1391.43 ms
Response P95	12547.07 ms	3534.10 ms	-9012.97 ms
RPS (Artillery)	108.00	251.33	+143.33
Error rate	0.00%	0.00%	0.00 %

Comparison: i18n v10 vs i18n-micro

Metric	i18n v10	i18n-micro	Difference
Max Memory	527.66 MB	474.45 MB	-53.21 MB
Avg Memory	418.26 MB	305.38 MB	-112.88 MB
Response Avg	857.43 ms	486.20 ms	-371.23 ms
Response P95	6702.60 ms	3534.10 ms	-3168.50 ms
RPS (Artillery)	161.33	251.33	+90.00
Error rate	0.00%	0.00%	0.00 %

Notes

• Shared profile: 4 locales × 2 pages × ~16.8k index leaves.
• Load: programmatic Artillery only — warm 6s@6 + main 60s@60, uncapped VU (historical YAML methodology). Paths from runtime profile — see scripts/src/perf/load.ts.
• Cool-downs (--cool fast): 0.2s post-build, 0.5s between runs, 0.5s between fixtures. Builds are forced each run so means are not diluted by cache hits.
• Artillery tables use HTTP Error rate; traffic charts also show vusers.failed (ETIMEDOUT under uncapped VU saturation) — not the same metric.
• Re-run day-to-day: pnpm test:performance (--load short --cool fast).
• Regenerate this page: pnpm -C scripts cli performance --load full --only all --runs 3.
• Flags: --locales N --keys K --only all|micro|i18n|plain --runs N --skip-load --cool fast|strict.