Amazon S3 Glacier
S3 Glacier is a set of archive storage classes inside S3 — the cheapest place on AWS to keep data you rarely touch but cannot delete: backups, compliance records, raw sensor data, archived video. You upload a normal S3 object and assign it a Glacier class directly or via a Lifecycle Rule.
Deep Archive costs about USD 0.00099 per GB-month — over 20 times cheaper than S3 Standard. The trade is retrieval latency: getting data back takes minutes to hours, so Glacier is for write-once, read-rarely data.
The Three Glacier Classes
Glacier Instant Retrieval reads in milliseconds, for rarely accessed data that must still be available instantly. Glacier Flexible Retrieval returns data in 1 minute to 12 hours, for backups where some delay is fine. Glacier Deep Archive takes 12 to 48 hours and is the cheapest, for 7+-year compliance retention.
Glacier Instant RetrievalGlacier Flexible RetrievalGlacier Deep ArchiveStorage price falls as retrieval time grows: roughly USD 0.004, 0.0036, and 0.00099 per GB-month respectively.
Retrieval Tiers and Minimums
For Flexible Retrieval and Deep Archive you pick a retrieval speed per request — Expedited (1–5 minutes, costliest), Standard (3–5 hours), or Bulk (cheapest, up to 12–48 hours). Retrievals are charged per GB and per request, so a large Expedited pull can be expensive.
Glacier classes also impose a minimum storage duration — 90 days for Flexible Retrieval, 180 for Deep Archive. Delete sooner and you still pay for the minimum, which makes Glacier wrong for short-lived data.
Glacier Deep Archive — the cheapest storage, when you can wait up to 48 hours to read. For long-term compliance archives.
Glacier Flexible/Instant — rarely-read data needing hours (Flexible) or instant (Instant) access at low storage cost.
S3 Standard / IA — data you read regularly. The retrieval fees and minimums make Glacier wrong for warm data.
- Putting frequently-read data in a Glacier class — retrieval fees quickly exceed what Standard or IA would have cost.
- Storing short-lived data in Glacier and deleting it early — the 90/180-day minimum-duration charge still applies.
- Requesting a large dataset back at Expedited speed without checking cost — bulk retrieval is far cheaper when you can wait.
- Using Flexible Retrieval for true cold data when Deep Archive is much cheaper and the extra delay is acceptable.
- Manually moving objects to Glacier instead of using age- or tag-based Lifecycle Rules.
- Skipping S3 Object Lock on compliance archives, leaving them deletable before their retention expires.
- Use Lifecycle Rules to transition objects to Glacier automatically by age.
- Use Deep Archive for true cold storage; reserve Flexible Retrieval for data you might need within hours.
- Plan retrievals and prefer Bulk speed when you can wait, to control retrieval cost.
- Tag objects by compliance category and filter Lifecycle Rules by tag.
- Use S3 Object Lock for WORM compliance archives that must not be deleted early.
Knowledge Check
What is the central trade-off of the Glacier storage classes?
- Much lower storage cost in exchange for retrieval latency of minutes to hours
- Higher object durability in exchange for a higher monthly per-GB storage cost
- Faster millisecond reads in exchange for noticeably lower durability
- Unlimited free retrievals in exchange for one flat monthly fee
Data must be readable within milliseconds but is accessed only a few times a year. Which class fits?
- Glacier Instant Retrieval — millisecond reads at low storage cost
- Glacier Deep Archive — the cheapest class, with millisecond reads
- Glacier Flexible Retrieval using the low-cost Bulk retrieval tier
- S3 Express One Zone for single-digit-millisecond access
Why is Glacier a poor fit for data you might delete within a month?
- The minimum storage duration (90 days Flexible, 180 Deep Archive) is billed even if you delete early
- Glacier flatly refuses to store any object smaller than 1 GB in size
- Glacier objects are strictly write-once and can never be deleted at all once they have been stored
- Glacier charges a higher per-GB monthly storage rate than S3 Standard
You need to restore a large Deep Archive dataset and cost matters more than speed. Which retrieval tier?
- Bulk — the cheapest tier, accepting up to 48 hours for Deep Archive
- Expedited — always reach for the single fastest retrieval tier on offer here
- Standard — it returns the data at no retrieval charge at all
- There is only a single retrieval tier available for Deep Archive
You got correct