system design cheat sheet

Question 1

CAP Theorem

Answer 1

• Consistency (C) → Both branches must always show the same account balance.
• Availability (A) → You can always withdraw money at any branch.
• Partition Tolerance (P) → Sometimes the network between branches fails.

Consistent when Partitioned (CP)

• Not available for writes
• However, data is consistent

Available when Partitioned (AP)

• Available for writes
• However, data is inconsistent

Question 2

Data Storage

Answer 2

Relational DB (RDS, Postgres, MySQL) → Strong consistency, joins, transactions.

NoSQL (MongoDB, DynamoDB, Cassandra) → Flexible schema, high write/read throughput, denormalization.

Object storage (S3, GCS) → Cheap, durable, good for files/blobs.

Time-series DB (InfluxDB, TimescaleDB) → Optimized for append-only, temporal queries.

Rule of thumb: Store metadata in DB, large blobs in S3.

Question 3

Caching

Answer 3

Where?

• Client-side → browser/app.
• CDN → static content close to users (CloudFront, Akamai).
• App-level → Redis, Memcached.
• DB query caching.

Eviction policies: LRU, LFU, TTL.

Question 4

Estimation (quick math for interviews)

Answer 4

Storage:
* 1 KB ≈ 1,000 bytes; 1 MB ≈ 1M bytes; 1 GB ≈ 1B bytes.
* Text row (id + timestamp + ~200 chars) ≈ 1 KB
* Photo (JPEG, compressed) ≈ 200 KB
* 1 minute of video (Full HD 1080p) ≈ 10 MB

Cache:
* 20% of total for 99% hit rate
* Each machine 1TB ram

RPS estimation:
* Lambda ≈ 1000; EC2 ≈ 500 (t3.medium) - 10,000 (c6i.4xlarge)
* DAU × requests/user ÷ seconds in a day.
* Example: 10M DAU × 20 req ÷ 86,400 ≈ 2,300 RPS.

Question 5

AWS Services Use Cases

Answer 5

• Storage → S3, DynamoDB, RDS
• Messaging → SNS, SQS, Kinesis
• Processing → Lambda, Step Functions, ECS
• Security/Identity → IAM, Cognito, KMS
• Analytics → Athena, Redshift, QuickSight

Question 6

When stuck

Answer 6

Storage, throughput, consistency

Question 7

Steps

Answer 7

1. Clarify (ask questions)
2. Define Requirements (functional and non-functional)
3. Identify Core Components (API Gateway, Lamdba, RDS, etc)
4. Draw a High-Level Architecture and Discuss Data Storage & Flow
5. Dive Into Critical Paths (ex: Feature Flags: flag evaluation is critical — focus on low-latency reads and cache invalidation)
6. Sketch 2–3 sample API endpoints
7. Handle Scaling & Reliability
8. Trade-offs & Alternatives
9. Wrap-Up & Questions