Module 12

Learning &
Revision

Spaced repetition flashcards, progress tracking, cheat sheets, and your personal notes.

0Due Today

0New

0Learning

0Mastered

—

Loading…

click to reveal answer

Answer

🎉

Session Complete!

All cards reviewed for today.

How SRS works: Rate each card after revealing the answer. Again = show again soon · Hard = shorter interval · Good = normal interval · Easy = longer interval. Cards you struggle with appear more often. Cards you know appear less often. Space = flip · → = next · ← = prev

Your Progress

Overall Completion

0% complete

🎯 Practice Interview

Architecture Cheat Sheet

Latency Numbers Every Engineer Must Know

L1 cache~0.5 ns — on-chip, essentially free

L2 cache~7 ns — 14× slower than L1

RAM access~100 ns — 200× slower than L1

SSD random read~150 µs — 150,000× slower than L1

Redis GET~0.5–2 ms

DB query (indexed)~5–20 ms

HDD seek + read~10 ms

Same-DC network~0.5 ms

US → EU (cross-ocean)~150 ms (speed of light + hops)

DNS lookup (uncached)~20–120 ms

TCP handshake1 RTT (e.g. 150 ms cross-continent)

TLS 1.3 handshake1 RTT (TLS 1.2 = 2 RTTs)

Capacity Estimation Rules of Thumb

Seconds/day86,400 ≈ 100K (use 100K to simplify)

1M req/day≈ 12 req/sec

1B req/day≈ 12,000 req/sec

1 KB × 1B records= 1 TB storage

1 MB × 1B records= 1 PB storage

Twitter: 100M tweets/day≈ 1,200 writes/sec

Average tweet size280 chars + metadata ≈ 500 bytes

Photo (compressed)~200 KB thumbnail, ~3 MB full

Database Selection Guide

PostgreSQL / MySQLACID, complex queries, JOINs, financial data, user accounts, inventory

RedisSessions, caching, rate limiting, leaderboards, pub/sub, real-time counters

CassandraWrite-heavy, time-series, event logs, masterless horizontal scale, no SPOF

MongoDBFlexible schema, nested documents, mobile/web backends, product catalogues

ElasticsearchFull-text search, log analytics, faceted filtering, autocomplete

Neo4jSocial graphs, fraud detection, recommendations, knowledge graphs

InfluxDB/TimescaleMetrics, IoT sensors, financial ticks, monitoring dashboards

BigQuery / RedshiftAnalytics, OLAP, data warehouse, large aggregations over historical data

CAP Theorem Quick Reference

CP systemsHBase, ZooKeeper, etcd, MongoDB (majority write concern) — returns error during partition

AP systemsDynamoDB, Cassandra, CouchDB, DNS — returns stale data during partition

The real choiceNetworks always partition → choose CP or AP; CA doesn't exist in distributed systems

PACELC extensionEven without partitions: tradeoff between Latency (fast) vs Consistency (coordinated)

Consistency Models (Weakest → Strongest)

EventualAll nodes agree eventually. Highest availability. DynamoDB default, DNS, Cassandra ONE.

Read-Your-WritesYou always see your own writes. Profile updates, shopping cart. Session guarantee.

CausalCausally related ops appear in order. MongoDB sessions, some distributed DBs.

LinearisableOps appear instantaneous. Strongest. etcd, Spanner, ZooKeeper. Highest latency.

Availability Nines

99% (2 nines)3.65 days downtime/year · 7.2 hrs/month

99.9% (3 nines)8.7 hours/year · 43.8 min/month

99.99% (4 nines)52 minutes/year · 4.4 min/month

99.999% (5 nines)5.2 minutes/year · 26 sec/month

Chained services rule3 services × 99.9% each = 99.7% combined. Chain degrades availability.

Key Algorithms Reference

Consistent HashingMinimises remapping when nodes change. ~1/N keys moved. DynamoDB, Cassandra, CDN.

RaftLeader election + log replication. Majority quorum. Used by etcd, CockroachDB.

Token BucketRate limiting. N tokens, replenish at rate R. Allows bursts up to bucket size.

Bloom FilterProbabilistic membership. No false negatives, possible false positives. Chrome safe browsing.

HyperLogLogApproximate cardinality. ~12KB any size. ~0.81% error. Redis PFADD/PFCOUNT.

LRU EvictionEvict least-recently-used item. O(1) with doubly-linked list + hashmap. Redis default.

Message Delivery Semantics

At-most-onceFire and forget. May lose messages. Use for: metrics, logging, analytics.

At-least-onceRetried until ACK. May duplicate. Consumers must be idempotent. Most common.

Exactly-onceKafka transactions + idempotent producer. ~20% throughput cost. Use sparingly.

Interview Framework (35-Minute Structure)

0–5 minClarify requirements: scope, scale, top features, constraints, SLAs

5–8 minCapacity estimation: users → QPS → storage → bandwidth → identify bottleneck

8–18 minHigh-level design: draw main components, get alignment before going deep

18–33 minDeep dive: focus on 2–3 hard components with interviewer guidance

33–35 minWrap up: tradeoffs, bottlenecks, what you'd do with more time

6-Week Interview Roadmap

Week 1 — Foundations + Networking

Complete Module 01 fully. CAP, ACID, scaling, rate limiting, circuit breakers — know them cold. Then Module 02 (Networking). The entire internet runs on these concepts.

Week 2 — Data Layer

Modules 03 (Databases) + 04 (Caching). 80% of system design problems reduce to data layer decisions. Know your indexing, NoSQL types, cache strategies. Play with the LRU simulator.

Week 3 — Async & Distribution

Modules 05 (Queues) + 06 (Distributed Systems). Kafka, delivery guarantees, Raft consensus. Use the Raft simulator. Understand sagas vs 2PC deeply — this is where candidates fail senior interviews.

Week 4 — Production Engineering

Modules 07–09: Scalability, DevOps, Security. CQRS, event sourcing, Docker/K8s, observability, OAuth/JWT. Show interviewers you think about production, not just whiteboard diagrams.

Week 5 — System Design Practice

Work through Module 10 + the Interview Simulator. Do each design yourself (35 min timer) before reviewing. Focus: URL shortener, YouTube, WhatsApp, Uber, Instagram — they cover the most patterns.

Week 6 — Mock Interviews + Refinement

Mock interviews daily. Practice talking while drawing. Flashcards every morning (20 min SRS session). Read one production engineering post per week (Netflix Tech Blog, Cloudflare Blog). You're ready.

Daily habits that compound: 20 minutes of SRS flashcards each morning. One timed system design sketch per day. One real engineering post-mortem per week. Interview intuition builds slowly — consistency beats cramming.

My Notes

Notes auto-save to your browser. They persist between sessions.

Learning &Revision

Session Complete!

Your Progress

Overall Completion

Architecture Cheat Sheet

Latency Numbers Every Engineer Must Know

Capacity Estimation Rules of Thumb

Database Selection Guide

CAP Theorem Quick Reference

Consistency Models (Weakest → Strongest)

Availability Nines

Key Algorithms Reference

Message Delivery Semantics

Interview Framework (35-Minute Structure)

6-Week Interview Roadmap

Week 1 — Foundations + Networking

Week 2 — Data Layer

Week 3 — Async & Distribution

Week 4 — Production Engineering

Week 5 — System Design Practice

Week 6 — Mock Interviews + Refinement

My Notes

Learning &
Revision