Design Zoom

Interview-Ready Approach

1) Clarify Scope and SLOs

• •Problem statement: Design Zoom's video conferencing platform - real-time video/audio, screen sharing, recording, and breakout rooms for 300 M+ daily participants.
• •Design for a peak load target around 80,000 RPS (including burst headroom).
• •Daily meeting participants: 300,000,000
• •Peak concurrent meetings: ~10,000,000
• •Max participants/meeting: 1,000
• •Video latency (end-to-end): < 150 ms
• •Audio latency: < 100 ms

2) Capacity Planning Method

• •Convert traffic and growth constraints into request rate, storage growth, and concurrency budgets.
• •Keep at least 2-3x safety margin per tier (ingress, compute, storage, async workers).
• •Reserve explicit latency budgets per hop so p95 can be defended in review.

3) Architecture Decisions

• •WebSockets: Use persistent connection gateways and decouple fanout via pub/sub or queues.
• •CDN: Serve static and cacheable content from edge and keep origin strictly for misses and dynamic requests.
• •Geo Distribution: Route users to nearest region/edge while keeping write-consistency boundaries explicit.
• •Databases: Define a clear system-of-record and design read/write paths separately before adding optimizations.
• •Monitoring: Instrument golden signals (latency, traffic, errors, saturation) per tier and per tenant/domain.
• •Auth: Centralize identity verification and keep authorization checks close to domain resources.

4) Reliability and Failure Strategy

• •Track connection churn, backpressure, and session resumption behavior.
• •Define cache keys and purge workflows before launch to avoid stale/global outages.
• •Design region failover and data residency controls as first-class requirements.
• •Use strong write constraints (transactions or conditional writes) and explicit backup/restore strategy.
• •Alert on user-impact SLOs, not only infrastructure metrics.

5) Validation Plan

• •Run one peak-load test, one dependency-degradation test, and one failover test.
• •Verify idempotency for all retried writes and async consumers.
• •Track user-facing SLOs first: p95 latency, error rate, and successful throughput.

6) Trade-offs to Call Out in Interviews

• •WebSockets: WebSockets reduce interaction latency but complicate scaling and state management.
• •CDN: Long TTL improves latency/cost; short TTL improves freshness.
• •Geo Distribution: Global latency improves, but cross-region consistency and operations become harder.
• •Databases: SQL gives stronger transactional guarantees; NoSQL often gives better write scaling and flexibility.
• •Monitoring: Deep observability speeds incident response but raises ingestion and tooling costs.

Practical Notes

• •SFU (Selective Forwarding Unit) per meeting: receives N streams, forwards up to N-1 to each participant. Only the active speaker + pinned videos are forwarded at high quality.
• •Simulcast: each sender encodes 3 quality layers. The SFU picks the right layer per receiver - low quality for thumbnail tiles, high for the active speaker.
• •Route meeting to the data center closest to the majority of participants. For geographically distributed meetings, use media relay bridges between data centers.

Why This Solution Works

Request path: The solution keeps ingress, service logic, and stateful dependencies separated so each layer can scale independently.

Reference flow: Mobile Clients -> DNS -> CDN Edge -> Load Balancer -> API Gateway -> Core Service -> Auth Service -> Primary NoSQL DB

Design strengths

• •Monitoring and logs are wired in from day one for rapid incident triage.
• •Security controls are enforced at ingress to protect downstream capacity.

Interview defense

• •This design makes bottlenecks explicit (ingress, core compute, persistence, async workers).
• •It supports progressive scaling without re-architecting the core request path.
• •It keeps correctness-sensitive state changes in durable systems while offloading background work asynchronously.