Live Capacity Planning Scenario

Overview

Calculate first, design second

Capacity planning interviews test whether you can translate a vague business requirement into concrete infrastructure decisions. The NALSD framework (Non-Abstract Large System Design) from Google's SRE writing gives the right sequence: calculate first, design second.

> [!IMPORTANT] A DAU number is useless until you turn it into a request pattern. Start with demand decomposition, not instance counts.

What interviewers look for

• Decomposition — DAU to requests per day to peak RPS
• Explicit headroom — GC overhead, safety margin, N+2 redundancy
• Stated assumptions — Read ratio, cache hit rate, no video
• Growth thinking — Where the architecture hits a cliff at 3x and 10x

Interview questions

• An interviewer says: "We expect 10 million daily active users on a new social feed service. Size the service." Walk through your approach.
• The interviewer follows up: "Traffic grows 3x in 6 months due to a viral feature. How do you handle it without rewriting the service?"

More SRE topics

• Identifying Scaling Bottlenecks
• Incident Command
• Incident Communications
• Incident Metrics
• Infrastructure as Code
• Launch Coordination
• Load Shedding
• Load Testing
• Logging Best Practices
• Modern Observability Stack
• Multi-cloud Active-Active
• Multi-Region Latency Spike Diagnosis