Back-of-envelope estimation
Estimate peak QPS, retained storage, memory and server count with explicit assumptions, consistent units and a failure reserve.
25 min8 steps
By the end you can
- Estimate peak request rate from daily users and actions
- Estimate retained replicated storage and hot-entry memory
- Calculate whole server count from a measured capacity and utilization target
- Identify assumptions and missing overhead that materially change a capacity estimate
Estimate to choose a design
Estimate to pick a design, not to get a precise bill. Registered users are not daily users.
Turn a day into a peak second
Divide a day by 86,400 s, then apply a stated peak factor.
Per day
24M
Average
278/s
Peak × 4
1,111/s
Loading question…
Retention multiplies daily ingestion
Retention multiplies daily ingestion. Count copies, indexes, backups and free space.
One copy
45 GB
Three copies
135 GB
before indexes and backups
Loading question…
Hot data includes more than payload
Hot data costs more than payload. Replicas come after that.
- Payload2 GB
- With keys and overhead (3×)6 GB
See caching before promising a hit rate.
Loading question…
Round up after applying usable capacity
Usable capacity = benchmark × target utilization. Round up.
Loading question…
Carry the estimate through the response path
Capacity uses the peak rate; daily volume uses the average. Include internal fan-out: 10 reads per request is 10,000 reads/s at 1,000 requests/s.
Payload
20 MB/s
Link
160 Mbit/s
Per day (constant)
1,728 GB
Ask what remains after a failure
Normal-state capacity is not an availability plan. Vary the inputs and find what breaks first.
Check your mastery
Score 80% or more. For each result, name one excluded overhead and one assumption to measure.
Mastery check
5 questions on fresh variants. One attempt each, no time limit. Score 80% or more to master the lesson.
Next: partitioning or bandwidth.
Answer this step to continue.