Design a Globally Distributed Rate Limiter Supporting 100M Requests Per Second
Technical Round 3 (System Design) · 0 years of experience
Staff Software Engineer at Google
The question
A way to think about it
Key Architectural Insight: Centralizing counters in a single global Redis creates unacceptable cross-region WAN latency (100ms+). The optimal design uses local in-memory token buckets synchronized asynchronously via distributed sliding-window consensus or token batching.
Ready to work through the answer?
Sign in for the complete explanation, examples, and follow-ups. Keep useful questions in your revision queue as you go.
Your question will be here when you get back.
92 people found this question helpful
Community approaches
Compare diverse candidate frameworks, code implementations, and trade-offs.
Sign in to compare approaches shared by the community and contribute your own.
Sign in to compare approachesDiscussion (0)
Ask questions, clarify prompts, or share interview variantsHave a question or a different perspective? Sign in to join the conversation.
Sign In