Drop-In Node to Bun Perfomance Evaluation

Related: Fooselo - Project

Overview

Following post captures performance measurements after migrating a Node.js project to Bun as a drop-in replacement. The goal is to evaluate real-world performance differences across common development tasks.

Goals

Measure and compare performance of:

  1. Install packages (cold & warm cache)
  2. Run unit tests
  3. Build client application
  4. Generate API client

Metrics Captured

  • Median runtime (seconds) and standard deviation (via repeated runs)
  • Wall-clock runtime per run
  • Peak memory (Maximum resident set size in MB)
  • Cache state (cold = cleared caches, warm = after first run)
  • Environment metadata (OS, CPU, RAM, Node version, pnpm version, Bun version, date/time)

Environment

Baseline (Node.js + pnpm)

  • Date: 2026-01-08
  • OS: Fedora 43 (Linux kernel 6.17.12-300.fc43.x86_64)
  • CPU: 11th Gen Intel(R) Core(TM) i7-11800H @ 2.30GHz (8 cores, 16 threads)
  • RAM: 31GB
  • Node.js: v25.0.0
  • pnpm: 10.20.0
  • Disk: NVMe SSD

After Migration (Bun)

  • Date: 2026-01-08
  • Bun: 1.3.5
  • (Other environment same as baseline)

Measurement Methodology

Tools Used

  • hyperfine: For stable repeated timings with statistical analysis
    • Install: sudo apt install hyperfine or cargo install hyperfine
  • GNU time: For memory usage measurements
    • Command: /usr/bin/time -v

Hyperfine Example

Process

  1. Run each command multiple times (warmup: 3, runs: 10)
  2. Capture median, mean, and standard deviation (hyperfine)
  3. Capture peak memory for at least one representative run (/usr/bin/time -v)
  4. Test both cold (cleared caches/node_modules) and warm (cached) scenarios
  5. Record exact command lines used

Commands Run

Use the automated script: scripts/bench/measure.sh

# Baseline (pnpm)
./scripts/bench/measure.sh "pnpm-install-cold" "rm -rf node_modules && pnpm install"
./scripts/bench/measure.sh "pnpm-install-warm" "pnpm install"
./scripts/bench/measure.sh "pnpm-test-client" "pnpm nx test client --skip-nx-cache"
./scripts/bench/measure.sh "pnpm-build-client" "pnpm nx build client"
./scripts/bench/measure.sh "pnpm-codegen" "pnpm nx run generated-api-client:codegen"

# After Bun migration
./scripts/bench/measure.sh "bun-install-cold" "rm -rf node_modules && bun install"
./scripts/bench/measure.sh "bun-install-warm" "bun install"
./scripts/bench/measure.sh "bun-test-client" "bunx nx test client --skip-nx-cache"
./scripts/bench/measure.sh "bun-build-client" "bunx nx build client"
./scripts/bench/measure.sh "bun-codegen" "bunx nx run generated-api-client:codegen"

Results

Package Installation

Variant Cache State Median (s) Stddev (s) Peak RSS (MB) Notes
pnpm cold 4.023 0.103 679.40 1544 packages
pnpm warm 0.837 0.015 216.61 Already up to date
bun cold 1.621 0.027 224.01 3085 packages installed
bun warm 0.178 0.006 205.94 No changes

Cold Install Speedup: 2.5x faster (59% time reduction)
Warm Install Speedup: 4.7x faster (79% time reduction)

Unit Tests (client)

Runtime Median (s) Stddev (s) Peak RSS (MB) Notes
pnpm/node 5.496 0.138 230.44 53 tests passed
bun 4.460 0.191 225.08 53 tests passed

Test Speedup: 1.2x faster (18% time reduction)

Build Client

Runtime Median (s) Stddev (s) Peak RSS (MB) Notes
pnpm/node 1.934 0.032 188.75 With Nx cache
bun 0.934 0.015 189.83 With Nx cache

Build Speedup: 2.1x faster (52% time reduction)

Generate API Client

Runtime Median (s) Stddev (s) Peak RSS (MB) Notes
pnpm/node 3.178 0.062 186.36 Orval codegen
bun 2.147 0.027 185.47 Orval codegen

Codegen Speedup: 1.5x faster (32% time reduction)

  • …

Follow-Up Steps

It is obvious that the main potential for reducing unit test execution time lies not in swapping the runtime. Bun offers a test runner, bun:test, as an alternative to Vitest or Jest. The next step is to migrate the project to bun:test and update the evaluation document.

The outlook is promising:

alt text

Sources: x/colinhacks

Bun Test Docs

References