MLRIT Performance Platform

A benchmarking platform that measures the quantitative impact of Redis caching and CDN optimizations on media-heavy web applications — built to demonstrate infrastructure engineering with real, measurable numbers.

What It Does

Upload media → serve through Redis cache → measure the latency difference in real time.

• First load: queries SQLite via Prisma, caches result in Upstash Redis (5-min TTL)
• Second load: reads from Redis — no DB query, measured <5ms round-trip
• Live Infrastructure HUD shows cache hit/miss state, Redis latency, and payload size per request
• Pressure Test Bench loads small images, large images, small videos, and 4K video to stress-test the cache layer

Measured Results

Metric	Before (Prisma)	After (Redis HIT)	Improvement
API latency	~50ms	~5ms	10× faster
Connection overhead	Per-request	Singleton	Zero leak risk
Cache hit rate	0%	~85% (repeated loads)	—

Stack

• Next.js (App Router) · TypeScript
• Upstash Redis — cache layer (REST API, singleton connection)
• Prisma + SQLite — database (dev)
• Cloudinary CDN — media delivery with automatic WebP/AVIF format negotiation
• Custom telemetry — X-Cache, X-Response-Time, X-Redis-Latency, X-Payload-Size response headers

Architecture

Browser → /api/media
           ├── Redis HIT  → return cached JSON (X-Cache: HIT, ~5ms)
           └── Redis MISS → Prisma query → cache in Redis → return (X-Cache: MISS, ~50ms)

/api/stats → ring buffer (last 100 requests) → cache hit rate, avg latency

Key Engineering Decisions

Singleton Redis connection — without this, 50 concurrent uploads = 50 open connections = server crash. Module-level const redis = makeRedis() runs once; Node.js module caching ensures every import gets the same instance.

Response header telemetry — instead of a separate logging service, latency data is stamped directly onto API responses (X-Redis-Latency, X-Response-Time). The client reads these headers to drive the live HUD — zero extra network round-trips.

Ring buffer for stats — only the last 100 requests are kept in memory. No database writes for metrics; no unbounded memory growth.

Run Locally

# Install dependencies
npm install

# Set up environment
cp .env.local.example .env.local
# Fill in: UPSTASH_REDIS_REST_URL, UPSTASH_REDIS_REST_TOKEN, DATABASE_URL, CLOUDINARY_CLOUD_NAME

# Seed the database (50 media items)
npx prisma db push
npx ts-node prisma/seed.ts

# Start dev server
npm run dev

Open http://localhost:3000/gallery

Environment Variables

See .env.local.example for all required variables.

Variable	Purpose
UPSTASH_REDIS_REST_URL	Upstash Redis endpoint
UPSTASH_REDIS_REST_TOKEN	Upstash auth token
DATABASE_URL	SQLite path (dev)
CLOUDINARY_CLOUD_NAME	CDN media delivery