Time
- 1 day
- 86.4K sec
- 1 month
- 2.6M sec
- 1 year
- 31.5M sec
System design, with the math left in
Turn vague scale into a design you can defend.
Practice back-of-the-envelope estimates for the systems interviewers ask about most. Change an assumption, watch the architecture move, and learn what each number is for.
- 9
- interview systems
- 45+
- worked formulas
- 0
- magic numbers
Live estimate
Peak traffic
231K req/s
Cache hit92%
Origin load18.5K/s
backend = peak × (1 − hit rate)
Interview signal The cache changes the database tier by an order of magnitude.
Traffic & QPSStorage growthNetwork bandwidthCache pressureShards & servers
01 Interactive scenario lab
Pick a system. Pressure-test the assumptions.
Every result is recalculated in your browser. No signup, no hidden model, no hand-waving.
Estimation conventionDecimal units · 86,400 sec/day · 3 replicas where stated
URL shortener
1
live
Set the assumptions
Edit any input to recalculate the design.
2
Read the capacity signals
Use these to choose components, not vendors.
3
Calculation walkthrough
Show the math, one decision at a time.
4
Reference architecture
Let the load shape the design.
Follow the primary request path left to right.
5
Scaling roadmap
Scale when the bottleneck earns it.
02 The reusable method
A five-minute framework for any scale question.
Use the same sequence whether you are designing a chat app, an object store, or a payment system. The numbers become a narrative for the architecture.
Useful precision beats false precision.Round aggressively, state what you ignored, and keep one safety factor visible.
- 01
Scope the behavior
Who uses it, how often, and which path must be correct or fast?
users × actions - 02
Convert to rates
Turn daily operations into average QPS, then make the peak explicit.
ops ÷ 86,400 × peak - 03
Price every byte
Estimate payload, retention, indexes, replication, and network direction.
items × bytes × time - 04
Find the hot path
Apply cache hit rates, fan-out, concurrency, or queueing amplification.
base load × amplification - 05
Draw the failure boundaries
Shard and isolate only where a measured load or correctness boundary calls for it.
capacity ÷ unit target
03 Pocket field guide
Numbers worth knowing before the interview.
Use these as starting assumptions, then replace them with measurements in a real system.
Bytes
- Short record
- 0.1–1 KB
- JSON response
- 1–100 KB
- Photo
- 0.2–5 MB
Safety factors
- Peak / average
- 3–10×
- Replication
- 3 copies
- Headroom
- 30–50%
Latency order
- Memory
- nanoseconds
- Same-region network
- milliseconds
- Cross-ocean
- 100+ ms