Performance Benchmarks

Independent, verifiable performance testing using industry-standard tools. MoroJS ships its own native C++ engine (@morojs/engine) as the default transport — measured here from the published npm packages. All benchmarks are reproducible with one command.

Moro Engine

Moro's own native C++ engine (@morojs/engine), the default transport, zero config.

Requests/sec102,409

Avg Latency0.9ms

Pipelined ×10572,053

Default · Peak Performance

• Native C++ engine, same Moro API

• Beats uWebSockets.js in both profiles

• Lowest latency (p99 1.6ms)

Clustering Mode

Built-in clustering with auto worker management.

Requests/sec103,971

Avg Latency0.9ms

Pipelined ×10584,016

Multi-Core

• Same native engine × auto workers

• SO_REUSEPORT load distribution

• Scales across all CPU cores

Node.js Path

Pure Node http server, the automatic fallback wherever a prebuilt engine binary isn't available.

Requests/sec68,163

Avg Latency1.4ms

Pipelined ×10119,570

Zero-Native Fallback

• No native binary required

• Automatic, nothing fails to boot

• Matches raw node:http throughput

Framework Comparison

Every framework measured on the same machine, same session, same tool no cross-sourcing. Two profiles: real-world (no pipelining, the headline how real HTTP clients behave) and pipelined ×10. Full data + methodology in the benchmark repo.

Framework	Req/sec real-world	Pipelined ×10	Latency	Runtime
MoroJS - clustered (engine × all cores)	103,971	584,016	0.9ms	Node 24
MoroJS - Moro Engine (default)	102,409	572,053	0.9ms	Node 24
MoroJS + uWebSockets.js (opt-in)	101,237	520,799	0.9ms	Node 24
Elysia	96,717	18,690	1.0ms	Bun
Fastify	69,375	117,069	1.4ms	Node 24
MoroJS - Node.js path (fallback)	68,163	119,570	1.4ms	Node 24
Elysia (Node adapter)	66,786	115,798	1.5ms	Node 24
Koa	60,924	93,903	1.6ms	Node 24
Hono	56,926	100,278	1.4ms	Node 24
Express	47,279	69,540	2.1ms	Node 24

vs uWebSockets.js+9.8%572k vs 521k · pipelined

vs Fastify+48%102k vs 69k · real-world

vs Express+117%102k vs 47k · real-world

vs Elysia (Bun)31×572k vs 19k · pipelined

Every row measured on one machine, one session (Mac Studio M2 Ultra, Node 24.11, wrk, best-of-3, both profiles). Reproduce any of it with node bench.js <target> --pipelined.

Key Findings

Moro Engine vs uWebSockets.js

Pipelined ×10 (through MoroJS)+9.8% faster

Real-world profileAhead in both

Bare engine vs bare uWS (peak)664k vs 648k

Moro Engine vs Node.js Path

Real-world Throughput+50% faster

Pipelined Throughput4.8x

Latency Reduction~36% lower

Detailed Performance Metrics

Moro Engine (default)

Real-world req/s: 102,409

Pipelined ×10 req/s: 572,053

Latency avg: 0.9ms

Latency p99: 1.6ms

RSS under load: 230 MB

Clustered (engine × all cores)

Real-world req/s: 103,971

Pipelined ×10 req/s: 584,016

Latency avg: 0.9ms

Latency p99: 1.6ms

Workers: auto (24)

Node.js Path (fallback)

Real-world req/s: 68,163

Pipelined ×10 req/s: 119,570

Latency avg: 1.4ms

Latency p99: 1.8ms

RSS under load: 216 MB

Testing Methodology

Tools & Environment

wrk (cross-checked with oha, bombardier)
Mac Studio M2 Ultra, idle machine
Node.js v24.11.0
Published npm packages, production config

Test Parameters

100 concurrent connections
40 second duration, best-of-3 per target
Both profiles: real-world + pipelined ×10
One server at a time, cooldowns between

Test Configurations

Moro Engine — default (port 3117)
Clustered mode (port 3111)
Node.js path (port 3110)
Boot sanity asserts the intended engine

Straight talk

These measure framework overhead, not your app. Your real numbers ride on your database, your I/O, and your business logic. What they do prove: MoroJS adds almost nothing on top of the transport it runs on. And every row is one command to reproduce - so don't take our word for it, run it.

Reproduce These Results

Clone & Setup

git clone https://github.com/Moro-JS/benchmark.git

cd benchmark && npm install

One command per target — no second terminal

# Moro Engine (the default) both profiles, best-of-3

node bench.js engine --pipelined

# Clustered / Node path / competitors

node bench.js cluster single fastify express koa --pipelined

# Everything, saved to a results-<timestamp>.md

npm run bench:save

How the harness works

Each run boots the server, waits for readiness, asserts the intended engine actually loaded, runs the load generator (a separate native process), samples memory, tears down, and cools down before the next target.

Quick sanity runs: npm run bench:quick (10s — never published). Raw baselines: node bench.js raw-engine raw-uws raw-node.