Choose the Right Shelf

What you'll learn

Every class except One Zone and Express One Zone keeps data across at least three Availability Zones; they differ in price, retrieval cost and retrieval time.
Infrequent-access and Glacier classes have minimum storage durations and per-GB retrieval charges.
Glacier Flexible Retrieval and Deep Archive need a restore before you can read the object.

Section 01

The shelves

Class	Best for	First byte	Minimum duration
`STANDARD`	Hot data read often	Milliseconds	—
`INTELLIGENT_TIERING`	Unknown or changing patterns	Milliseconds (optional archive tiers need a restore)	—
`STANDARD_IA`	Read about once a month	Milliseconds	30 days
`ONEZONE_IA`	Re-creatable data, one AZ	Milliseconds	30 days
`GLACIER_IR`	Read a few times a year, needed instantly	Milliseconds	90 days
`GLACIER`	Archives you can wait minutes to hours for	Restore: 1–5 min to 12 h	90 days
`DEEP_ARCHIVE`	Compliance archives, rarely if ever read	Restore: up to 12 h (48 h bulk)	180 days

GLACIER is the API name for S3 Glacier Flexible Retrieval. Infrequent-access and Glacier Instant Retrieval bill a minimum of 128 KB per object.

Section 02

Minimums and retrieval fees

Cheaper storage comes with conditions. Delete a Standard-IA object after 10 days and you still pay for 30. Read a gigabyte from Standard-IA or Glacier Instant Retrieval and you pay a retrieval fee per GB. Small objects are billed as if they were 128 KB.

So the right class depends on the access pattern, not just the price per GB-month. For hot, small cards, STANDARD is often the cheapest in total.

Section 03

Restores

Objects in Glacier Flexible Retrieval and Deep Archive aren't directly readable: GetObject returns InvalidObjectState. Call RestoreObject to make a temporary copy readable for a number of days, and wait for the restore to finish.

Section 04

Intelligent-Tiering

If you can't predict access, Intelligent-Tiering moves each object between tiers for you: to an infrequent tier after 30 days without access and to an archive instant-access tier after 90, and back to frequent the moment it's read. It charges a small monitoring fee per object (objects under 128 KB aren't monitored or moved).

Practice

Your turn

Tournament battle replays are watched a few times a year, must start playing the moment someone clicks, and are kept for years. Choose the storage class that costs least for that pattern.

Set StorageClass on the upload.

// @ts-nocheck
import { S3Client, PutObjectCommand } from "@aws-sdk/client-s3";
const client = new S3Client({ region: "ap-south-1" });
async function archiveReplay(replay) {
  await client.send(
    new PutObjectCommand({
      Bucket: "hoenn-pokedex-media",
      Key: "replays/2026/hoenn-league-final.mp4",
      Body: replay,
      ContentType: "video/mp4",
      // ⬇️ Pick a storage class. (Leaving it out means STANDARD.)
    })
  );
}

// @ts-nocheck
import { S3Client, PutObjectCommand } from "@aws-sdk/client-s3";

const client = new S3Client({ region: "ap-south-1" });

async function archiveReplay(replay) {
  await client.send(
    new PutObjectCommand({
      Bucket: "hoenn-pokedex-media",
      Key: "replays/2026/hoenn-league-final.mp4",
      Body: replay,
      ContentType: "video/mp4",
      // ⬇️ Pick a storage class. (Leaving it out means STANDARD.)

    })
  );
}

Ctrl/⌘ Enter to run · simulated, no AWS callsSimulated — no AWS calls

Check

Check your understanding

Quick Quiz · one question, as many tries as you need.

Raw sensor dumps can be regenerated from the source, are read about once a month, and must be available instantly. Which class fits best?

Choose one answer.