1. Requirements (5 min) → Functional + Non-functional + Constraints
2. Estimation (3 min) → QPS, storage, bandwidth
3. High-Level Design (10 min) → Core components + data flow
4. Deep Dive (20 min) → Critical paths, bottlenecks, trade-offs
5. Wrap Up (7 min) → Monitoring, failure modes, scaling
Resource Latency Throughput
L1 cache 0.5 ns - L2 cache 7 ns - RAM 100 ns - SSD 150 μs 100K IOPS HDD 10 ms 200 IOPS Same DC network 0.5 ms - Cross-continent 150 ms -
Time Seconds
1 day 86,400 1 month 2.6M 1 year 31.5M
1 million users × 10 reads/day = ~116 QPS
1 KB × 1M/day = 1 GB/day = ~365 GB/year
Pattern When Trade-off
Vertical Simple, early stage Single point of failure Horizontal Need redundancy/throughput Complexity, consistency Database Sharding Write-heavy, large data Cross-shard queries hard Read Replicas Read-heavy Replication lag Caching Repeated reads Cache invalidation CDN Static content, global Cost, cache staleness Async (Queues) Bursty traffic Eventual consistency
Strategy Write Path Read Path Consistency
Cache-aside App → DB, invalidate cache App → Cache → DB Eventual Write-through App → Cache → DB App → Cache Strong Write-back App → Cache → (async) DB App → Cache Risk of loss Refresh-ahead Background refresh App → Cache Near real-time
L4 (Transport): TCP/UDP level, fast, no content inspection
L7 (Application): HTTP level, content-aware, routing by path/header
Algorithms: Round Robin, Weighted, Least Connections, IP Hash, Consistent Hash
Health Checks: Active (periodic probe) + Passive (error monitoring)
99.9% = 8.76 hours/year downtime
99.99% = 52.6 min/year downtime
99.999% = 5.26 min/year downtime
Serial: A_total = A1 × A2 × A3
Parallel: A_total = 1 - (1-A1)(1-A2)
Pattern Latency Consistency Use Case
Strong High Linearizable Banking, inventory Eventual Low Converges Social media, feeds Read-your-writes Low Per-user User profiles Monotonic reads Low No going back Timelines
Need Choice
ACID, relations PostgreSQL, MySQL Massive writes, wide columns Cassandra, HBase Document store MongoDB, DynamoDB Key-value (fast) Redis, Memcached Graph Neo4j, Dgraph Search Elasticsearch Time series InfluxDB, TimescaleDB Object/blob S3, GCS
CQRS: Separate read/write models
Event Sourcing: Store events, not state
Saga: Distributed transactions with compensating actions
Circuit Breaker: Fail fast on downstream failures
Bulkhead: Isolate critical resources
Rate Limiting: Token bucket, sliding window, leaky bucket
REST: Resources, HTTP verbs, stateless
gRPC: Protobuf, streaming, internal services
GraphQL: Client-driven queries, single endpoint
Versioning: URL path (/v1/), header, query param
Pagination: Cursor-based (stable), Offset-based (simple)
Rate Limiting: Per-user, per-IP, token bucket