GitHub - pshenok/server-survival: Tower defense game that teaches cloud architecture. Build infrastructure, survive traffic, learn scaling.

Server Survival is an interactive 3D simulation game where you play as a Cloud Architect. Your mission is to build and scale a resilient cloud infrastructure to handle increasing traffic loads while fighting off DDoS attacks, managing your budget, and keeping your services healthy.

Learn cloud by playing:

🏭 New: Datacenter Survival — the sister game. This one teaches the logical layer of the cloud; that one teaches the physical layer it runs on: power chains, heat, cooling, PUE. Play it here.

Game Modes

• Survival — the core experience: survive as long as possible against escalating traffic, DDoS spikes, random events, and service degradation.
• Campaign — 25 hand-crafted levels across 5 chapters (Basics → Optimization → Defense & Mastery → Production Readiness → The AI Wave). Each level teaches one architecture concept with pre-built diagrams, objectives, and a debrief.
• Sandbox — a free-play lab: any budget, any traffic mix, no game over. Experiment with all 26 services.

What It Teaches

Every mechanic is a real cloud concept in miniature:

• Observability — place a Monitoring node to unlock a live metrics dashboard (RPS, error rate, latency sparklines) and alerts. Level 15 makes you fly blind without it first.
• Auto-scaling & cold start — Compute fleets scale out at 70% utilization and back in at 30%, but new instances need warmup time (Containers even more so).
• Queue-depth scaling — a queue-fed fleet scales on queue pressure, not just CPU, so backlogs trigger scale-out.
• Circuit breaking & retries — overloaded downstreams trip a breaker (50% error rate over a rolling window), traffic is retried with backoff, and SPOFs get flagged.
• Multi-region failover — GeoDNS splits traffic across regional stacks; region-outage events show why one region is never enough.
• Inference serving & batching — the GPU Cluster batches INFERENCE requests (up to 8/12/16 by tier) into one job that amortizes a fixed per-batch cost, so profit is all utilization: a full batch prints money, a half-fed GPU bleeds its $60/min upkeep.
• Model cold starts & quality — a GPU loads its model for 12/20/30s before serving anything, and every tier upgrade reloads it; bigger models also answer better (10% → 4% → 1% bad-answer risk, −0.5 reputation per bad answer).
• Inference SLOs — the Inference Gateway buffers for warming or busy GPUs but expires anything older than its 6s deadline: a stale generation is failed honestly, never served.
• Power as the constraint — the grid carries 8 kW, every GPU draws 6, and a Substation adds 6: the fleet's ceiling is bought in watts before it is bought in dollars.
• OLTP vs OLAP — the Data Warehouse takes analytics writes cheaply but refuses realtime reads; the Relational DB does the opposite trade.
• Failure-reason badges — every dropped request pops a badge on the node that dropped it, saying why (no route, capacity, read-only replica, search-only index...).

How to Play

Objective

Survive as long as possible! Manage your Budget ($), Reputation (%), and Service Health.

• Earn Money by successfully processing legitimate traffic requests.
• Lose Reputation if requests fail or if malicious traffic slips through.
• Maintain Health - Services degrade under load and need repairs.
• Game Over if Reputation hits 0% or you go bankrupt ($-1000).

Traffic Types

Infrastructure & Services

Build your architecture using the toolbar. All 26 services are organized into five category tabs — the same taxonomy as the table below. Each service has a cost, capacity, and per-minute upkeep:

Scoring & Economy

Upkeep & Cost Scaling

• Base Upkeep: Each service has per-minute upkeep costs
• Upkeep Scaling: Costs increase 1x to 2x over 10 minutes
• Repair Costs: 15% of service cost to manually repair
• Auto-Repair: +10% upkeep overhead when enabled

Survival Mode

The core experience - survive as long as possible against escalating traffic with constant intervention required:

• RPS Acceleration - Traffic multiplies at time milestones (×1.3 at 1min → ×4.0 at 10min)
• Random Events - Cost spikes, capacity drops, traffic bursts, service outages every 15-45 seconds
• Traffic Shifts - Traffic patterns change every 40 seconds
• DDoS Spikes - 50% malicious traffic waves every 45 seconds
• Service Degradation - Services lose health under load, require repairs
• Health bars, event indicator, finances panel, auto-repair toggle, game-over analysis

Campaign Mode

25 levels across 5 chapters, each one a scenario built around a single lesson:

1. Basics (levels 1-3) — your first server, storage, and edge caching with a CDN.
2. Optimization (levels 4-10) — caching, queues, read scaling, search, NoSQL, rate limiting, serverless vs compute.
3. Defense & Mastery (levels 11-14) — defense in depth, high availability, cost crunches, and Black Friday.
4. Production Readiness (levels 15-20) — observability, auto-scaling, node failures, dead-letter queues, fan-out, and two-region failover.
5. The AI Wave (levels 21-25) — GPU inference: model cold starts, batch-fill economics, SLO deadlines, the power wall, and a capstone that has to end in the black.

Levels come with pre-built starting architectures, primary and bonus objectives, speedrun stars, and a debrief tip. Later chapters throw forced outages — including a whole region going dark — at your build.

Sandbox Mode

A fully customizable testing environment for experimenting with any architecture:

No game over in Sandbox - experiment freely!

Share Your Architecture

Built something you're proud of? The share panel exports your architecture two ways:

• PNG export — a clean image of your diagram to post anywhere.
• Shareable URL — your whole build encoded in a ?arch= link; anyone who opens it gets your architecture loaded and ready to run.

Controls

• Left Click: Select tools, place services, and connect nodes.
• Right Click + Drag: Pan the camera.
• Scroll: Zoom in and out.
• WASD / Arrows: Move camera (pan) when zoomed in.
• ESC: Open main menu and pause game. Press again or click Resume to close menu (stays paused).
• Camera Reset: Press R to reset the camera position.
• Birds-Eye View: Press T to switch between isometric and top-down view.
• Hide HUD: Press H to toggle UI panels.
• Category Tabs: The toolbar groups services into five tabs — Front Door, Compute, Data, Async, Ops. Keys 1-5 switch between them.
• Connect Tool: Click two nodes to create a connection (flow direction matters!).
  • Valid Flows: Internet -> (GeoDNS/Firewall/CDN/API Gateway) -> Load Balancer -> Queue -> Compute -> Cache -> (Search Engine/Read Replica/SQL DB/NoSQL DB/Storage)
• Delete Tool: Remove services to recover 50% of the cost.
• Time Controls: Pause, Play (1x), and Fast Forward (3x).

Strategy Tips

1. Block Attacks First: Always place a Firewall immediately connected to the Internet. Malicious leaks destroy reputation fast (-5 per leak) and cost $50 per breach.
2. Use CDN for Static Content: Connect Internet -> CDN -> Storage. The CDN handles 95% of static traffic cheaply!
3. Watch Service Health: Damaged services have reduced capacity. Click to repair or enable Auto-Repair.
4. Scale for Traffic Surges: RPS multiplies at milestones - prepare before ×2.0 at 3 minutes!
5. Balance Income vs Upkeep: Start lean, scale as income grows. Over-provisioning leads to bankruptcy.
6. Use Cache Wisely: Reduces database load significantly for READ requests.
7. Buffer with Queue: Queue helps survive traffic burst events without dropping requests — and a queue-fed Compute fleet auto-scales on queue depth.
8. React to Events: Watch the event bar - cost spikes mean hold off on purchases, traffic bursts mean ensure capacity.
9. API Gateway for Graceful Degradation: Throttled requests only lose -0.2 reputation (vs -1.0 for failures). Great for surviving traffic spikes!
10. Split DB Traffic with NoSQL: Route READ/WRITE to NoSQL (faster, cheaper) and keep the Relational DB for SEARCH queries only.
11. Search Engine for SEARCH Storms: SEARCH is the heaviest traffic type (2.5x weight). A dedicated Search Engine processes it 3x faster than SQL DB.
12. Read Replica for READ offloading: Under "API Heavy" or "Read Heavy" shifts, a Read Replica prevents your main DB from drowning in READ requests.
13. Serverless Function = pay-per-use: Great for spiky traffic, but monitor finances — per-request cost ($0.03) adds up fast at high RPS.
14. Place Monitoring early: The metrics dashboard shows error rates and latency before the failure badges start raining.
15. Read the failure badges: Every dropped request tells you why it died — the badge is the lesson.

Tech Stack

• Core: Vanilla JavaScript (native ES modules — an explicit module graph under src/, no bundler)
• Rendering: Three.js for 3D visualization.
• Styling: Tailwind CSS for the glassmorphism UI.
• Build: No build step required! The game is served raw from this repo by GitHub Pages.

Getting Started

To play, click PLAY NOW. No install, no build, no account.

To run your own copy — a fork, or a change you just made — serve the folder instead of opening the file. The game is native ES modules, and browsers fetch module scripts in CORS mode, so a file:// origin is blocked and double-clicking index.html gets you the UI shell over an empty canvas. Any static server works, and one is already on your machine:

git clone https://github.com/pshenok/server-survival.git
cd server-survival
python3 -m http.server 8000    # then open http://localhost:8000

For Contributors

There is still zero build step — the dev tooling is optional and for contributors only:

• npm install once, then npm run check runs ESLint + the full Vitest suite (593 tests).
• CI runs the same check on every PR.
• The code is native ESM: game.js plus focused modules under src/ (sim/, core/, ui/, campaign/, persistence/, input/).

See CONTRIBUTING.md for the workflow.

Community

Join our Discord server to discuss strategies and share your high scores: Join Discord

Built with code and chaos.