Feature Store Size Calculator

Work out feature store size instantly with clear inputs, formula shown and shareable results.

About Feature Store Size Calculator

A feature store has two tiers with very different sizes. The online store holds only the latest value per entity, so it is entities times features times bytes and must fit in low-latency storage. The offline store keeps a point-in-time-correct history for training, multiplying that footprint by the retention window, which is why 90 days of history costs 90 times the online tier.

Formula

Feature store sizing

online = entities x features x bytes; offline = online x history days; daily growth = one online snapshot

Worked example

With the default values below, Feature Store Size Calculator returns Online store size of 17.88 GiB. Change any input above and every figure here is recalculated in your browser.

Inputs used

Entitiesminimum 1: 20,000,000
Features per entityminimum 1: 120
Bytes per feature valueminimum 1 bytes: 8 bytes
Days of history in the offline storeminimum 1: 90

Results

Online store size: 17.88 GiB
Offline store size: 1.57 TiB
Bytes per entity: 960 B
Offline store growth per day: 17.88 GiB

Step by step

1Values used
2Feature store sizing
3Online store size= 19,200,000,000
4Offline store size= 1,728,000,000,000
5Bytes per entity= 960 B
6Offline store growth per day= 19,200,000,000

Assumptions and limits

The result is derived only from the 4 inputs listed above. Anything not among them is not part of the calculation.
In particular, this calculator does not take inflation, tax, fees or charges or compounding frequency into account.
Figures are rounded for display to whole numbers, so totals built from several rounded parts can differ slightly from the unrounded arithmetic.

How these calculators are built and tested is described on the Methodology page. If a figure looks wrong, please report it.

Frequently Asked Questions

Why keep so much history offline?

Training sets must be built with the feature values as they were at each event time. Without history, backfilling a training set introduces look-ahead leakage.

How can the online tier be shrunk?

Serve only the features models actually consume, use compact types instead of doubles and strings, and expire entities that have been inactive beyond a sensible window.