S3 storage classes and retrieval
Lesson 1 of 4 in Lifecycle and Tiering.
Every S3 storage class holds the same bytes with the same durability design. What differs is the bargain, and each class strikes it in three dimensions at once: price per gigabyte-month, what a read costs in money and in time, and how long you are billed for whether you keep the object or not.
Learn the bargain and the classes stop being a list to memorise.
S3 Standard — no minimum duration, no retrieval charge, milliseconds. The default, and the right answer for anything actively queried.
S3 Standard-IA and S3 One Zone-IA — cheaper per gigabyte, a per-gigabyte retrieval charge, still milliseconds, and a 30-day minimum. One Zone-IA additionally stores in a single Availability Zone, which is the only class that lowers the durability expectation: lose the zone, lose the data. It is for data you can regenerate.
S3 Glacier Instant Retrieval — archive pricing with milliseconds access and a 90-day minimum. The class people forget exists, and the only archive class that needs no restore.
S3 Glacier Flexible Retrieval — 90-day minimum, and an object cannot be read at all until you restore it: Expedited in 1–5 minutes, Standard in 3–5 hours, or Bulk in 5–12 hours free of retrieval charge.
S3 Glacier Deep Archive — the cheapest, a 180-day minimum, and restore in about 12 hours Standard or within 48 hours Bulk.
The sentence that decides half of all storage-class questions is therefore: does the data have to be readable on demand? If yes, the two restore-requiring classes are out however cheap they are — and if a stem says "queryable by auditors without a restore", it has told you so explicitly.
| Class | Read latency | Minimum billed duration | Availability Zones | Choose it when |
|---|---|---|---|---|
S3 Standard | milliseconds | none | ≥3 | the data is actively queried or written |
S3 Intelligent-Tiering | milliseconds in the automatic tiers | none (a small monitoring charge per object) | ≥3 | the access pattern is unknown or changing, and objects are ≥128 KB |
S3 Standard-IA | milliseconds | 30 days | ≥3 | read a few times a month, and a retrieval charge is acceptable |
S3 One Zone-IA | milliseconds | 30 days | 1 — the data is lost with the zone | the data can be regenerated; never for an authoritative copy |
S3 Glacier Instant Retrieval | milliseconds | 90 days | ≥3 | archive economics with on-demand reads — queried once or twice a quarter |
S3 Glacier Flexible Retrieval | restore first: 1–5 min, 3–5 h, or 5–12 h free | 90 days | ≥3 | backup and recovery sets where a wait of hours is planned for |
S3 Glacier Deep Archive | restore first: ~12 h, or within 48 h bulk | 180 days | ≥3 | compliance retention nobody expects to read — the cheapest gigabyte there is |
S3 Intelligent-Tiering is the answer to a different question: not "how cold is this data" but "I do not know, and it changes".
Objects start in Frequent Access, drop to Infrequent Access after 30 consecutive days without access, and then to Archive Instant Access after 90 — all three with milliseconds retrieval and no retrieval charge — and any access promotes the object straight back to Frequent Access. Two further tiers can be opted into: Archive Access from 90 days and Deep Archive Access from 180, which do require a restore and therefore reintroduce the problem the automatic tiers avoid.
It charges a small monitoring and automation fee per object, and there is no minimum storage duration. That fee is what decides whether it fits, and it produces the one trap worth memorising: objects smaller than 128 KB are not monitored and are always billed at Frequent Access rates. A bucket of 400 million tiny objects on Intelligent-Tiering pays the monitoring fee and receives none of the tiering, which is strictly worse than leaving them in Standard.
The comparison to make is therefore between Intelligent-Tiering and an explicit lifecycle policy. Explicit policy is cheaper when you genuinely know the pattern — raw partitions are hot for 30 days, warm for a year, cold forever — because there is no monitoring fee and you control the dates. Intelligent-Tiering wins when access is irregular or unpredictable, when the data is read occasionally in ways nobody can schedule, and specifically when an explicit policy would risk tiering something into a class a reader cannot use.
A horizontal bar chart of illustrative storage price per terabyte-month for seven Amazon S3 storage classes, shaped like typical US East pricing rather than quoted from the price list. S3 Standard is about 23 dollars per terabyte-month with milliseconds access and no retrieval charge. S3 Standard-Infrequent Access is about 12.50 dollars with milliseconds access and a per-gigabyte retrieval charge. S3 One Zone-Infrequent Access is about 10 dollars with the same access profile but stored in a single Availability Zone. S3 Glacier Instant Retrieval is about 4 dollars with milliseconds access and a ninety-day minimum. S3 Glacier Flexible Retrieval is about 3.60 dollars but cannot be read without a restore taking between one minute and twelve hours. S3 Glacier Deep Archive is about 1 dollar and needs a restore of about twelve hours or up to forty-eight for bulk. The teaching point is that the span from top to bottom is roughly twenty-three to one in storage price, while the span in time to first byte is from twenty milliseconds to forty-eight hours — so the saving is always bought with access.
Key terms: S3 storage class, Minimum storage duration, S3 Intelligent-Tiering, Restore request
Interactive checkpoint quiz (2 questions) — open this page in a browser to take it.