• Ask before design
• Keep it simple (people prefer simple system over complex one)
• Working solution, not perfect solution
• Analysis > Solution. Tradeoffs!! Bottlenecks

The entire interview is around 45mins. Deducting 10 mins for questions, we have ~35mins for design.

• Functional requirements (Ask question; clarify workflow)
• Non-functional requirements
  • CAP (AP or CP?)
    • Partition tolerant
    • Availability (system downtime)
      • 5 = 99.999% = 5.26 minutes / year (AWS standard; highly available)
      • Data Replication (for disaster recovery)
      • Load balancing (global → different region; local → exclude dead servers)
    • Consistency (eventual consistency; read your write; transaction)
  • Latency.
    • Good DB choice (e.g. Bigtable for thumbnails)
    • Cache
    • CDN
    • Geographically distributed servers
  • Scalability
    • Horizontal Scaling
    • Load Balancer
    • Data Sharding
  • Reliability / Durability (data loss; redundancy)
    • Data Replication
    • Message Queue (producer and consumer can fail independently without affecting the others)
  • Fault-tolerance
    • Failover
    • Data Replication
  • Read-heavy or Write-heavy?
  • Observability
  • Maintainability
    • Micro-service (Scale independently)
    • Message Queue
• API Design
  • Don't have to talk about the details of the API signatures.
• DB Design (Tradeoff)
  • Entity
  • Relationship
    • 1:1: single table OR 2 tables with the same primary key
    • 1:m: foreign key
    • m:n: middle (relationship) table
  • DB Choice
• Estimate (Account for spikes, growth rate and redundancy)
  • Assumptions:
    • Total Users (for total storage, etc), DAU (for QPS)
    • Rate (10 photo uploads / s)
    • File size
  • Considerations:
    • QPS (Read; Write)
    • Storage
    • Bandwidth

Limit the above to 10~15mins. Save time for the high-level design.

• High-level design
• Detailed design
  • Ask the interviewer which part he/she is interested in
• Check if we achieved the non-functional requirements
  • Fault-tolerance:
    • Failover (active-active; active-passive)
    • Data Replication (Single-leader; Multi-leader; Leaderless)
  • Scaling:
    • Load-balancer (round-robin; minimum connection; minimum load)
    • Data Sharding
  • Reliability/Redundancy/Availability: Data Replication (Single-leader; Multi-leader; Leaderless)