Estimation Worksheet (Modern, 2026 Numbers)

Fill in the blanks for the system you are designing. The point is an order of magnitude, not precision.

Latency numbers worth memorizing (modern)

Operation	Rough time
L1 cache reference	~1 ns
Main memory reference	~100 ns
Redis GET (same DC)	~0.1–0.5 ms
SSD random read	~50–150 µs
Relational database indexed point read	~0.2–1 ms
Intra-DC network round trip	~0.5 ms
Cross-region round trip (US↔EU)	~80–150 ms
Kafka produce (acked)	~1–5 ms

Powers of two

Power	Exact	Approx	Name
10	1,024	1 thousand	KB
20	1,048,576	1 million	MB
30	~1.07e9	1 billion	GB
40	~1.10e12	1 trillion	TB

Capacity template

Daily active users (DAU):          __________
Actions / user / day:              __________
Read : Write ratio:                __________ : 1

Writes/day  = DAU * actions * write_fraction = __________
Writes/sec  = writes/day / 86,400            = __________
Peak QPS    = avg QPS * 2..10 (peak factor)  = __________

Avg record size:                   __________ bytes
Storage/day = writes/day * size              = __________
Storage/5yr = storage/day * 365 * 5          = __________

Bandwidth   = peak QPS * payload size        = __________

Sanity checks (do you actually need machinery?)

• Writes/sec well under ~10k and storage under ~10 TB? A single tuned relational database (+ read replica, replication and failover) likely suffices — do not shard yet (when not to scale).
• Read-heavy with a hot set? A cache (caching) before a shard (partitioning and sharding).
• 1M+ msgs/s of events? Now a log (Kafka, event streaming) earns its place.