S3 storage classes explained: which one should you use?

Every object in Amazon S3 has a storage class, and it’s the biggest lever on your S3 bill. The classes trade a lower storage price for something else: slower retrieval, fees when you read, minimum storage periods, or fewer copies. Choosing well comes down to two questions about your data:

• How often will it be read? Daily, monthly, quarterly, or almost never?
• How fast do you need it when you do? Milliseconds, minutes, or hours?

What every class has in common

All S3 storage classes are designed for 99.999999999% durability, “eleven nines”: it’s vanishingly unlikely you’ll lose an object. Most keep copies across at least three Availability Zones. The exceptions are the two “One Zone” classes, which keep data in a single Availability Zone, so if that zone is destroyed, the data is gone.

What changes from class to class is availability, retrieval time and how you’re charged.

The classes, from hot to cold

S3 Standard

The default, for data read often.

• Retrieval: milliseconds, no retrieval fee.
• Minimums: none.
• Use for: websites, content, data being actively processed.

S3 Intelligent-Tiering

For data whose access pattern you don’t know or that changes. S3 watches each object and moves it to a cheaper tier when it isn’t used: an Infrequent Access tier after 30 days, an Archive Instant Access tier after 90, and optional archive tiers beyond that.

• Retrieval: milliseconds for the default tiers, no retrieval fees; objects move back to the frequent tier when they’re read.
• Cost: a small monthly monitoring fee per object. Objects under 128 KB aren’t monitored and stay in the frequent tier.
• Use for: data lakes, user uploads, anything where you can’t predict reads.

S3 Standard-IA (Infrequent Access)

For data read about once a month that must still come back instantly when it is.

• Retrieval: milliseconds, with a per-GB retrieval fee.
• Minimums: 30 days of storage and 128 KB per object are always billed.
• Use for: backups, disaster recovery copies, older content that’s occasionally requested.

S3 One Zone-IA

Like Standard-IA, but cheaper, because it keeps data in one Availability Zone.

• Retrieval: milliseconds, with a per-GB retrieval fee.
• Minimums: 30 days, 128 KB.
• Use for: data you can re-create if lost, such as thumbnails or a secondary copy of a backup.

S3 Glacier Instant Retrieval

Archive prices for data read about once a quarter, still returned in milliseconds.

• Retrieval: milliseconds, with a higher per-GB retrieval fee than Standard-IA.
• Minimums: 90 days, 128 KB.
• Use for: medical images, news archives, anything old that someone may suddenly need right away.

S3 Glacier Flexible Retrieval

For archives you read once or twice a year and can wait for. Objects must be restored before you can read them.

• Retrieval: expedited in 1–5 minutes, standard in 3–5 hours, bulk in 5–12 hours (bulk retrievals are free).
• Minimums: 90 days.
• Use for: backups and archives with no urgency.

S3 Glacier Deep Archive

The cheapest S3 storage class, for data you keep for years and hope never to read.

• Retrieval: standard within 12 hours, bulk within 48 hours.
• Minimums: 180 days.
• Use for: compliance records kept for 7–10 years, a replacement for tape.

S3 Express One Zone

The odd one out: not cheaper, but faster. It stores data in a single Availability Zone, in directory buckets, with single-digit-millisecond access, for request-heavy workloads such as machine learning training and analytics. Storage costs more than S3 Standard; requests cost less.

Side by side

Moving data automatically: lifecycle rules

You rarely set classes object by object. An S3 Lifecycle configuration on the bucket does it for you, with two kinds of action:

• Transitions move objects to a cheaper class as they age: for example, to Standard-IA after 30 days, Glacier Flexible Retrieval after 90 and Deep Archive after a year.
• Expirations delete objects (or old versions) once you no longer need them.

Two things to watch:

• Lifecycle rules don’t move objects into Standard-IA or One Zone-IA until they’re at least 30 days old.
• Minimum storage periods still apply. Delete or move an object out of Glacier Flexible Retrieval after 30 days and you’re billed for the full 90.

If you can’t predict access well enough to write rules, Intelligent-Tiering is the lifecycle rule that writes itself.

Exam traps

• “Access patterns are unknown or change over time.” S3 Intelligent-Tiering.
• “Accessed about once a month, must be available immediately.” S3 Standard-IA.
• “Data can be re-created; lowest cost for infrequent access.” S3 One Zone-IA. Not for the only copy of anything important.
• “Accessed once a quarter, needs millisecond retrieval.” S3 Glacier Instant Retrieval.
• “Archive; retrieval within minutes is occasionally needed.” Glacier Flexible Retrieval with expedited retrieval.
• “Keep for 7 years for compliance, lowest cost, retrieval within 48 hours is fine.” S3 Glacier Deep Archive.
• “Millions of tiny objects moved to Standard-IA, and the bill went up.” Each object is billed as at least 128 KB.
• “Move objects to cheaper storage as they age.” S3 Lifecycle transitions.

For the full details, see the Amazon S3 User Guide on storage classes, S3 Intelligent-Tiering, the S3 Glacier storage classes, retrieval options, managing the object lifecycle and S3 Express One Zone.