Google Maps / Navigation

Google Maps is a global navigation and mapping platform used by over 2 billion people monthly. A user types an origin and a destination and receives turn-by-turn directions, a live-rendered map, and an estimated arrival time that updates continuously as they drive. Behind that interaction are three fundamentally different subsystems: a road-network graph traversal engine that finds the fastest path through hundreds of millions of nodes in milliseconds; a tile pipeline that pre-renders the planet’s geography into billions of image tiles stored in object storage and delivered via a global CDN; and a real-time traffic system that ingests a continuous stream of GPS breadcrumbs from every navigating device, aggregates them into edge-speed estimates, and propagates weight updates back to the routing engine within seconds.

Each sub-problem is independently hard at continental scale, and all three must be tightly integrated so rerouting feels instantaneous when a traffic jam appears ahead.

Functional Requirements

1. Route computation: Given an origin and destination (lat/lng or place name), return a turn-by-turn route with distance, ETA, and manoeuvre steps. Support fastest-time and shortest-distance modes.
2. Map tile serving: Render and serve raster or vector map tiles at zoom levels 0–20. Tiles load as users pan and zoom.
3. Real-time traffic: Ingest GPS probe data from navigating users, detect congestion, update ETAs within seconds, and reroute when a significantly faster path exists.
4. Route alternatives: Return 2–3 ranked alternative routes alongside the primary.
5. POI search: Return nearby points of interest given a text query and current location.
6. ETA estimation: Predict arrival time using historical traffic patterns, live conditions, and ML-based corrections.
7. Offline maps: Allow users to download a region for offline navigation with incremental delta sync on reconnect.

Out of Scope

• Street View imagery capture and serving (a separate camera-crawl and storage pipeline).
• Public transit routing (requires GTFS feed ingestion — a distinct service).
• 3D building rendering and satellite imagery processing.
• Ride-sharing or driver dispatch integration.
• Business listing management and user reviews.
• Voice synthesis for turn instructions (handled client-side from the route payload).

Non-Functional Requirements

• Scale: 2 B registered users; 50 M daily active navigating users; 20 M route requests/day; 50 B map tile requests/day; ~2 M GPS probe events/second from concurrent navigating sessions.
• Latency: Route computation p99 < 500 ms (city-level routes < 100 ms); map tile serving p99 < 50 ms globally; GPS probe ingest p99 < 200 ms.
• Availability: 99.99 % for tile serving and active turn-by-turn navigation (users are driving and depend on it); 99.9 % for route computation.
• Consistency: Traffic edge weights eventually consistent across all routing servers within ≤ 30 s of observation. Route results may use traffic data up to 30 s stale — acceptable for navigation.
• Durability: Road graph and tile data — zero loss tolerated; geo data is expensive to recreate. GPS probe data — best-effort; individual probe loss has no user-visible effect.
• Privacy: GPS probes anonymised at ingest (stripped of device ID); raw user location not persisted beyond the active navigation session.

Terminology