Benchmarks

Summary

ergo is benchmarked against five Node.js frameworks across nine scenarios that exercise different stages of the Fast Fail pipeline. Each scenario isolates a specific concern — routing, auth, body parsing, compression, conditional requests, rate limiting — to measure where frameworks spend time and how ergo’s design decisions affect throughput and latency.

Run date: 2026-03-22  |  ergo version: 0.1.0  |  Node.js: 22 (Alpine)

Overall Rankings (by average requests per second)

Scenario	1st	2nd	3rd	4th	5th	6th
Baseline GET	Hono (37,183)	Koa (36,841)	node:http (36,722)	ergo (36,472)	Fastify (35,049)	Express (18,581)
Param GET	ergo (38,798)	Hono (37,487)	node:http (36,493)	Koa (36,309)	Fastify (36,190)	Express (17,539)
Auth GET	ergo (36,437)	node:http (35,041)	Hono (34,500)	Koa (33,424)	Fastify (32,051)	Express (18,197)
JSON POST	ergo (34,389)	Koa (33,786)	node:http (33,008)	Fastify (31,993)	Hono (30,811)	Express (14,885)
Full Pipeline	ergo (34,688)	Koa (34,014)	node:http (33,107)	Fastify (31,752)	Hono (29,795)	Express (14,704)
Concurrency Ramp	Koa (30,249)	ergo (30,190)	Hono (28,969)	node:http (27,930)	Fastify (25,133)	Express (14,312)
Production Stack	node:http (18,008)	ergo (7,386)	Koa (6,564)	Express (4,834)	Hono (4,725)	Fastify (4,456)
Conditional GET	Koa (40,626)	Fastify (38,824)	node:http (37,859)	Hono (37,198)	ergo (36,612)	Express (17,953)
Rate-Limit Flood	Fastify (35,315)	node:http (34,626)	Hono (34,115)	Koa (33,046)	ergo (15,430)	Express (14,035)

node:httpExpressFastifyHonoKoaergo

Average requests per second by framework across all 9 scenarios. Higher is better.

ergo Highlights

• Param GET: #1 — 38,798 RPS (106% of node:http, 107% of Fastify)
• Auth GET: #1 — 36,437 RPS (104% of node:http, 114% of Fastify)
• JSON POST: #1 — 34,389 RPS (104% of node:http, 107% of Fastify)
• Full Pipeline: #1 — 34,688 RPS (105% of node:http, 109% of Fastify)
• Concurrency Ramp: #2 — 30,190 RPS (108% of node:http, 120% of Fastify)
• Production Stack: #2 — 7,386 RPS (166% of Fastify, 156% of Hono)
• Baseline GET: #4 — 36,472 RPS (99% of node:http, 104% of Fastify)
• Conditional GET: #5 — 36,612 RPS (97% of node:http, 94% of Fastify)
• Rate-Limit Flood: #5 — 15,430 RPS (45% of node:http, 44% of Fastify)

ergo ranks #1 in the four core pipeline scenarios (Param GET, Auth GET, JSON POST, Full Pipeline) — the scenarios that exercise the Fast Fail design most directly. In the Production Stack scenario with compression, CORS, content negotiation, and timeout middleware, ergo leads all full-featured frameworks at 166% of Fastify’s throughput.

node:httpExpressFastifyHonoKoaergo

Throughput degradation as middleware complexity increases from Baseline to Production Stack. Flatter lines indicate lower middleware overhead.

---

Per-Scenario Results

All values are the mean across 3 trial runs.

01 — Baseline GET

GET /ping — routing overhead only. Measures the framework’s minimum per-request cost with no middleware.

Framework	Avg RPS	p50 (ms)	p95 (ms)	p99 (ms)	Avg Latency (ms)	Mem Peak
Hono	37,183	0.88	2.19	4.64	1.06	29.1 MB
Koa	36,841	0.88	2.25	4.89	1.07	32.7 MB
node:http	36,722	0.89	2.21	4.80	1.08	22.4 MB
ergo	36,472	0.88	2.27	5.10	1.09	31.3 MB
Fastify	35,049	0.91	2.48	5.42	1.13	22.8 MB
Express	18,581	2.10	5.18	7.03	2.14	31.8 MB

02 — Param GET

GET /users/:id?fields=name — parameterized route with query string parsing. Exercises Stage 1 (Negotiation) — URL and query parameter handling.

Framework	Avg RPS	p50 (ms)	p95 (ms)	p99 (ms)	Avg Latency (ms)	Mem Peak
ergo	38,798	0.85	2.08	4.46	1.02	30.3 MB
Hono	37,487	0.87	2.17	4.71	1.05	28.6 MB
node:http	36,493	0.90	2.24	4.75	1.08	21.6 MB
Koa	36,309	0.89	2.29	4.96	1.09	33.5 MB
Fastify	36,190	0.87	2.34	5.21	1.09	30.8 MB
Express	17,539	2.25	5.38	7.26	2.27	32.8 MB

03 — Auth GET

GET /auth/users/:id with Bearer token — authenticated request. Exercises Stage 2 (Authorization) — credential extraction and token verification.

Framework	Avg RPS	p50 (ms)	p95 (ms)	p99 (ms)	Avg Latency (ms)	Mem Peak
ergo	36,437	0.90	2.23	4.68	1.08	29.4 MB
node:http	35,041	0.94	2.33	4.85	1.13	20.4 MB
Hono	34,500	0.95	2.37	5.00	1.15	29.3 MB
Koa	33,424	0.97	2.53	5.26	1.18	33.1 MB
Fastify	32,051	0.95	3.02	6.17	1.24	25.1 MB
Express	18,197	2.15	5.26	7.17	2.18	31.1 MB

04 — JSON POST

POST /users with JSON body — body parsing. Exercises Stage 3 (Validation) — request body stream reading and JSON deserialization.

Framework	Avg RPS	p50 (ms)	p95 (ms)	p99 (ms)	Avg Latency (ms)	Mem Peak
ergo	34,389	0.94	2.43	5.32	1.15	31.0 MB
Koa	33,786	0.96	2.48	5.22	1.17	33.4 MB
node:http	33,008	0.98	2.47	5.41	1.20	27.5 MB
Fastify	31,993	0.96	3.39	5.89	1.24	33.7 MB
Hono	30,811	1.11	2.92	7.57	1.28	37.2 MB
Express	14,885	2.70	6.03	7.83	2.67	33.5 MB

05 — Full Pipeline

POST /auth/users with Bearer token, JSON body, and AJV validation — exercises all four Fast Fail stages in sequence.

Framework	Avg RPS	p50 (ms)	p95 (ms)	p99 (ms)	Avg Latency (ms)	Mem Peak
ergo	34,688	0.94	2.35	5.04	1.14	29.7 MB
Koa	34,014	0.95	2.44	5.27	1.16	33.2 MB
node:http	33,107	0.99	2.49	5.14	1.19	28.4 MB
Fastify	31,752	0.98	3.38	5.69	1.25	35.8 MB
Hono	29,795	1.18	2.95	7.72	1.33	38.2 MB
Express	14,704	2.72	6.33	8.08	2.71	35.2 MB

06 — Concurrency Ramp

POST /auth/users with a ramp from 10 to 500 virtual users — measures throughput and latency under increasing concurrency to find the saturation point.

Framework	Avg RPS	p50 (ms)	p95 (ms)	p99 (ms)	Avg Latency (ms)	Mem Peak
Koa	30,249	2.27	19.05	32.18	5.40	33.7 MB
ergo	30,190	2.29	19.04	32.25	5.40	31.3 MB
Hono	28,969	2.57	19.08	37.17	5.63	39.0 MB
node:http	27,930	2.58	20.39	37.11	5.85	28.4 MB
Fastify	25,133	3.04	21.73	40.68	6.49	37.8 MB
Express	14,312	5.87	39.28	49.04	11.42	49.5 MB

07 — Production Stack

POST /stack/auth/users with CORS, content negotiation, timeout, auth, body parsing, AJV validation, and gzip compression — a realistic production middleware stack. The delta from Scenario 05 isolates the cost of additional middleware layers.

Framework	Avg RPS	p50 (ms)	p95 (ms)	p99 (ms)	Avg Latency (ms)	Mem Peak
node:http	18,008	1.96	5.14	9.14	2.20	23.1 MB
ergo	7,386	4.01	12.38	14.62	5.40	30.5 MB
Koa	6,564	4.50	14.42	17.67	6.08	32.6 MB
Express	4,834	8.54	14.17	17.22	8.26	28.0 MB
Hono	4,725	7.72	17.72	21.31	8.45	34.6 MB
Fastify	4,456	8.45	17.35	22.86	8.97	33.7 MB

08 — Conditional GET (ETag)

GET /cached/users/:id with If-None-Match — measures ETag generation and 304 Not Modified short-circuiting. Tests how efficiently frameworks skip serialization and compression when the resource hasn’t changed.

Framework	Avg RPS	p50 (ms)	p95 (ms)	p99 (ms)	Avg Latency (ms)	Mem Peak
Koa	40,626	0.81	1.99	4.26	0.97	32.7 MB
Fastify	38,824	0.84	2.09	4.61	1.02	23.7 MB
node:http	37,859	0.86	2.15	4.60	1.04	21.2 MB
Hono	37,198	0.87	2.23	4.89	1.06	29.1 MB
ergo	36,612	0.87	2.31	5.07	1.08	31.2 MB
Express	17,953	2.16	5.35	7.37	2.22	31.6 MB

09 — Rate-Limit Flood

POST /rate-limited/users under sustained flood — rate limit set to 50 requests per 10-second window with 50 concurrent VUs. The vast majority of requests are rejected with 429. Tests the Fast Fail principle: rejected requests should be cheap because the rate limiter runs in Stage 1 before body parsing, auth, and validation.

Framework	Avg RPS	p50 (ms)	p95 (ms)	p99 (ms)	Avg Latency (ms)	Mem Peak
Fastify	35,315	0.93	2.28	5.00	1.12	32.1 MB
node:http	34,626	0.94	2.35	5.06	1.14	24.7 MB
Hono	34,115	0.95	2.46	5.20	1.16	28.4 MB
Koa	33,046	0.95	2.65	6.14	1.21	32.4 MB
ergo	15,430	0.81	2.73	5.98	2.58	31.1 MB
Express	14,035	2.85	6.59	8.51	2.84	32.4 MB

Latency Distribution

p50p95p99

Latency distribution across four key scenarios. Tighter dot clusters indicate more predictable response times. Connecting lines show the p50-to-p99 spread.

Memory Footprint

node:httpExpressFastifyHonoKoaergo

Peak memory usage (MB) by framework across all scenarios. Lower is better.

---

Methodology

Docker Isolation

Each server runs in its own Docker container, one at a time (sequential, never concurrent), to eliminate inter-container CPU contention.

Resource	Server Container	k6 Container
CPU	--cpuset-cpus="0"	--cpuset-cpus="1"
Memory	--memory=512m	--memory=1g
Network	Docker bridge (bench)	Docker bridge (bench)
Node.js	node:22-alpine	—
NODE_ENV	production	—

All servers listen on port 3000 and implement identical request/response logic — the only variable is the framework.

Load Profile

k6 staged virtual users per scenario:

Phase	Duration	Virtual Users	Purpose
Warmup	30s	0 → 50	JIT warm, connection pools established
Sustain	60s	50	Measurement window
Ramp-down	10s	50 → 0	Graceful drain

Three trial runs per scenario × framework combination (162 total runs). The report computes the mean, standard deviation, and coefficient of variation (CoV) across trials to quantify run-to-run consistency.

Validation Parity

All frameworks use AJV JSON Schema validation for request body validation (Scenarios 05 and 07). Fastify uses its built-in AJV integration; all others compile schemas at startup. This ensures validation overhead is consistent across the suite.

---

Framework Versions

Framework	Version	Key Dependencies
node:http	Node.js 22 (Alpine)	ajv 8.18.0, etag 1.8.1
Express	5.2.1	compression 1.8.1, cors 2.8.6, express-rate-limit 8.3.1
Fastify	5.8.2	@fastify/compress 8.3.1, @fastify/cors 11.2.0, @fastify/etag 6.1.0, @fastify/rate-limit 10.3.0
Hono	4.12.8	@hono/node-server 1.19.11, ajv 8.18.0, etag 1.8.1
Koa	3.1.2	@koa/cors 5.0.0, @koa/router 15.4.0, koa-bodyparser 4.4.1, koa-compress 5.2.1
ergo	0.1.0	ergo-router 0.1.0

---

Statistical Rigor

Each framework × scenario combination is run 3 times. The report computes the coefficient of variation (CoV = stddev / mean × 100%) for both RPS and p99 latency to quantify measurement noise.

CoV Range	Verdict
< 3%	Excellent
3–5%	Acceptable
5–10%	Noisy
> 10%	Unreliable

This Run

• Average CoV across all 54 combinations: 1.8%
• Noisy (CoV ≥ 5%): 5 / 54 combinations
• Unreliable (CoV ≥ 10%): 0 / 54 combinations
• Overall verdict: CLEAN RUN — results are highly reliable

The noisy combinations were:

Framework	Scenario	CoV (RPS)	CoV (p99)
Fastify	Baseline GET	6.3%	12.9%
Fastify	Auth GET	8.6%	18.0%
ergo	JSON POST	5.2%	13.2%
Express	Conditional GET	5.2%	4.2%
Koa	Rate-Limit Flood	9.2%	18.6%

The Fastify variability in Scenarios 01 and 03 is consistent with Fastify’s JIT-sensitive startup behavior under short warmup windows.

---

Environment

OS:       Darwin 25.3.0 arm64
CPU:      Apple M4 Pro
Cores:    14 (12 performance + 2 efficiency)
RAM:      48 GB
Docker:   Docker Desktop 4.42.0 (Engine 28.1.1)
Node.js:  22 (pinned via node:22-alpine)
k6:       grafana/k6:latest

---

Reproduce It

The complete benchmark suite — scenarios, server implementations, Docker orchestration, and report generator — is committed to the ergo repository for full auditability.

cd benchmarks
chmod +x run.sh
./run.sh                    # ~2.5 hours for 162 runs
node generate-report.js     # produces results/report.md

Prerequisites: Docker Desktop (or Docker Engine) with multi-core CPU pinning support, 2+ CPU cores available, ~4 GB free RAM.

Synthetic benchmarks

These benchmarks are synthetic and hardware-dependent. Results will vary across machines, Docker runtimes, and OS configurations. The scenarios are designed to test ergo’s specific design decisions (Fast Fail ordering, synchronous middleware fast-path, return-value pipeline model) rather than to declare a universal “fastest framework.”