Skip to main contentSkip to user menuSkip to navigation
Interactive design lab

Capacity Examples Workbench

What is a capacity estimate?

A capacity estimate converts product demand into explicit traffic, compute, storage, and network assumptions that can be challenged.

Step 1

Model

Start with an illustrative workload, tune its demand and provisioning assumptions independently, then inject a spike or failure.

Step 2

Observe

Read the formulas, find the limiting budget, and turn an overloaded result into a concrete scale-out or load-shedding decision.

Step 3

Challenge

Compare planned load with a launch spike, slow dependency, and regional loss; then restore at least 30% service headroom.

Capacity estimation workbench

Turn product demand into an operating envelope

Start from an illustrative workload, expose every multiplier, then test whether the provisioned fleet survives a bad day.

1. Choose a workload

Use an example as a starting hypothesis

Illustrative assumptions, not company facts

Demand breakdown

Realtime messaging

Partition by conversation or recipient while keeping retries idempotent.

Invariant: daily requests = actions x requests per action
Daily actions1BProduct demand
Daily requests2.4B2.4x amplification
Planned peak88.9K/s3.2x average

3. Provision the envelope

Capacity and durability assumptions

This is the supply loop. Change fleet throughput or data durability without changing product demand.

Instances, pods, workers, or accelerator replicas.

80

A benchmark result at acceptable tail latency, not a vendor maximum.

1.8K/s

How long the stored payload remains online.

30 days

Primary plus replicas; compression and indexes are excluded.

3x

4. Challenge the plan

What happens on a bad day?

Select a mode to change demand, available capacity, and latency together.

Healthy reserve

62% of available service capacity

55.1K/s remains for variance; validate the 1.8K RPS-per-unit assumption with a representative load test.

Estimated p9570 msqueueing heuristic
Demand 88.9K/sAvailable 144K/s
Compute budget
71 units required

peak RPS / (unit RPS x 70% target x available fleet)

Storage budget
135 TB

actions x payload x 30 days x 3 copies

Network budget
444.4 Mbps

1.5 TB/day of logical user payload

Gateway88.9K/s
Message service80 units
Durable log135 TB
Decimal units are used. Compute follows amplified internal requests; storage and network follow one payload per user action. Protocol overhead, indexes, compression, backups, and growth are excluded.
No quiz questions available
Could not load questions file
Spotted an issue or have a better explanation? This page is open source.Edit on GitHub·Suggest an improvement