Last updated: 4th March 2026
Version: 1.0
Authors: Nicolas TORO
Status: Done
{.is-success}

---

Benchmark

This repository holds all the prototype benchmarks produced during the Ascension tech-stack selection phase. Each sub-project is a minimal, self-contained implementation used to validate a technology choice before committing to it in the main monorepo.

---

Table of Contents

• Benchmark
  • Table of Contents
  • Context
  • Repository Structure
  • Backend / Database Benchmarks
    • Rust + Axum + PostgreSQL
    • Go + Gin + MariaDB
    • Backend Decision
  • Mobile Benchmarks
    • Flutter
    • React Native (Expo)
    • Mobile Decision
  • What Each Benchmark Tests
  • How to Run the Benchmarks
    • Backend — Rust + PostgreSQL
    • Backend — Go + MariaDB
    • Mobile — Flutter
    • Mobile — React Native
  • Next Steps

---

Context

Before writing a single line of production code, the team ran focused benchmarks on the two most debated stack choices:

1. Backend + database: which API framework and database combination handles our workload best?
2. Mobile: which cross-platform framework gives us the best camera/video experience on Android and iOS?

Each benchmark is intentionally minimal — it implements only the features that matter for the decision criterion (performance, DX, ecosystem fit, video pipeline), not a full application.

---

Repository Structure

benchmark/
├── backend-db/
│   ├── rust-postgresql/   # Axum + SeaORM + PostgreSQL
│   └── go-mariadb/        # Gin + GORM + MariaDB
└── mobile/
    ├── flutter/           # Flutter + Dart
    └── react-native/      # React Native + Expo (TypeScript)

---

Backend / Database Benchmarks

Both backends implement the same minimal feature set so results are directly comparable:

• POST /register — user sign-up with bcrypt password hashing
• POST /login — authentication returning a JWT token
• POST /upload — authenticated video upload endpoint (JWT-protected)

Rust + Axum + PostgreSQL

Item	Detail
Framework	Axum 0.8
ORM	SeaORM 1.x
Database	PostgreSQL 16 (Docker)
Async runtime	Tokio
Password hashing	bcrypt
Observability	tracing + tower-http TraceLayer

The Rust implementation is the most bare-bones of the two: the HTTP layer, ORM, and database schema are all wired up in a single main.rs with no framework boilerplate. The database table is created at startup if it does not exist.

Go + Gin + MariaDB

Item	Detail
Framework	Gin 1.11
ORM	GORM + MariaDB driver
Database	MariaDB (via MySQL driver)
Auth	JWT (golang-jwt/jwt v5)
Password hashing	golang.org/x/crypto/bcrypt
Architecture	Layered: handler → service → repository

The Go implementation uses a full layered architecture (transport / service / repository / model) to evaluate how the language handles a realistic project structure, not just raw performance.

Backend Decision

✅ Chosen: Rust + Axum + PostgreSQL

Key reasons:

• Performance: Rust's zero-cost abstractions and async Tokio runtime outperform Go/Gin in raw throughput and memory usage at every concurrency level tested.
• Type safety: SeaORM's compile-time checked queries eliminate an entire class of runtime errors that showed up during the Go prototype.
• PostgreSQL over MariaDB: better JSON support, full ACID compliance, and stronger ecosystem for the analytical queries the AI pipeline requires.
• Consistency: the AI worker already uses Python; adding Go would introduce a third language. Rust gives us systems-level control without the complexity of a third runtime.

---

Mobile Benchmarks

Both mobile prototypes implement the same user-facing flow:

1. Home screen — entry point, navigation trigger.
2. Camera screen — request camera + microphone permissions, record a video (max 30 s).
3. Preview screen — play back the recorded video, display file size and duration, simulate an upload.

Flutter

Item	Detail
Language	Dart
Navigation	go_router ^14
Camera	camera ^0.10.5
Video playback	video_player ^2.8.2
Permissions	permission_handler ^11.3
HTTP	http ^1.2
Min SDK	Dart ≥ 3.0

The Flutter prototype targets Android and iOS from a single codebase. The camera initialisation flow, permission handling, and video metadata extraction (path, size, duration) are all implemented from scratch using native Flutter packages.

React Native (Expo)

Item	Detail
Language	TypeScript
Framework	React Native 0.81 + Expo ~54
Navigation	React Navigation 7 (native stack)
Camera	expo-camera ~17
Video playback	expo-av ~16 / expo-video ~3
HTTP	axios ^1.13
Target	Android, iOS, Web (Expo Go)

The React Native prototype uses the Expo managed workflow for rapid iteration. The same three-screen flow (Home → Camera → Preview) is implemented in TypeScript with typed navigation.

Mobile Decision

✅ Chosen: Flutter

Key reasons:

• Camera reliability: the Flutter camera package gave consistent behaviour across Android and iOS during testing. The Expo camera stack required more workarounds for permission flows and codec compatibility.
• Performance: Flutter's compiled Dart code rendered the camera preview and video playback noticeably more smoothly than the React Native JS bridge on the same physical devices.
• Single language: the team already uses Dart for Flutter; React Native would require maintaining TypeScript tooling separately.
• Video pipeline fit: Flutter's direct FFI access will be essential when integrating the AI video analysis pipeline. The React Native bridge adds latency that is unacceptable for real-time feedback.

---

What Each Benchmark Tests

Benchmark	Key criteria evaluated
Rust + PostgreSQL	Throughput under concurrent load, type-safe ORM, Docker setup time
Go + MariaDB	Architecture scalability, JWT middleware, GORM ergonomics
Flutter	Camera init reliability, permission UX, video metadata, cross-platform parity
React Native	Expo DX, camera codec compatibility, navigation type safety, bundle size

---

How to Run the Benchmarks

Backend — Rust + PostgreSQL

cd benchmark/backend-db/rust-postgresql

# Start PostgreSQL
docker compose up -d

# Run the API (reads DATABASE_URL from .env)
cargo run

The server starts on http://localhost:3000.

Available routes:

GET  /           → health check
POST /register   → { "email": "...", "password": "..." }
POST /login      → { "email": "...", "password": "..." }

Backend — Go + MariaDB

cd benchmark/backend-db/go-mariadb

# Copy and fill in the environment file
cp .env.example .env

# Run
go run ./cmd/app

Mobile — Flutter

cd benchmark/mobile/flutter

flutter pub get
flutter run

Requires a physical device or emulator with camera support. Android and iOS are both supported.

Mobile — React Native

cd benchmark/mobile/react-native

npm install
npx expo start

Scan the QR code with Expo Go, or press a / i to open the Android emulator / iOS simulator.

---

Next Steps

The benchmarks informed the final stack for the production monorepo:

• Backend: Rust + Axum + PostgreSQL → apps/server/
• Mobile: Flutter + Dart → apps/mobile/

These benchmark sub-projects are archived — they are kept for reference and are not maintained or updated alongside the production code.