[System Design] Uber RAMEN: How Ride-Hailing Apps Push Real-Time Notifications to Millions of Devices

Short summary

Uber's RAMEN system maintains persistent gRPC bidirectional streams to millions of driver apps, delivering ride offers in under 100ms instead of polling every 3 seconds. Using binary framing and connection multiplexing, it eliminates 99% of wasted empty requests and dramatically reduces battery drain. Cluster management via Apache Helix handles stateful routing and automatic failover when servers fail.

• •gRPC bidirectional streams deliver <100ms latency vs polling's 0-3s delay
• •Eliminates 99% of wasted empty requests and reduces battery drain
• •Apache Helix manages stateful routing and automatic rebalancing

Generated with AI, which can make mistakes.