Skip to content

SYSTEM Cited by 1 source

Amazon S3 Glacier

S3 Glacier is the cold-storage class family within Amazon S3 for archival data — objects stored for long durations with infrequent access. Three named tiers: Glacier Instant Retrieval, Glacier Flexible Retrieval, and Glacier Deep Archive, each with progressively lower per-GB storage cost and progressively longer retrieval times / higher retrieval fees.

Defining trade

Glacier classes extract the deepest per-GB storage savings of any S3 storage tier — but only if access is rare and access volume matches expectations. Two structural taxes erode the savings if access turns out to be more frequent than expected:

  1. Minimum storage durations — early deletion before the minimum period charges as if the object had been stored for the full minimum (90 days for Instant / Flexible Retrieval; 180 days for Deep Archive).
  2. Retrieval fees — per-GB read fees on top of standard request fees; for some classes, a tiered restore latency vs cost trade.

These two taxes combine into the cold-storage minimum- duration tax — the structural failure mode that punishes uncertain access.

Yelp's framing (2026-05-21)

"This is in contrast to cold storage classes (e.g., S3 Glacier) that impose minimum storage durations and retrieval fees that can negate savings if you access data more than you expected to."

Yelp's data-platform team uses Glacier as the comparison point that motivates S3 Intelligent Tiering as the default for unpredictable access patterns. IT's 81% savings at 90-day no-access "approaches the cost of S3 Glacier!" — capturing most of the cold-storage savings without the asymmetric tax.

(Source: sources/2026-05-21-yelp-how-partition-access-visualizations-reduced-our-data-lake-s3-cost-by-33)

When Glacier still wins

Glacier remains the right answer when:

  • Access is provably rare — confirmed retention requirements (compliance, regulatory) with no analytical-read use case.
  • Retrieval cost is predictable — known restore cadence for known use cases (e.g. annual audit pulls).
  • Storage volume is large enough that the deeper per-GB savings matter more than the retrieval-fee risk — exabyte-scale archives where IT's 81% no-access savings still leave material savings on the table.

For ambiguous-access workloads (the bulk of analytics-table data), Yelp's framing is that IT is the better bet.

Tier structure (named in canonical AWS product literature)

  • Glacier Instant Retrieval — milliseconds to read; 90-day minimum.
  • Glacier Flexible Retrieval — minutes to hours to read; 90-day minimum; multiple restore-speed options.
  • Glacier Deep Archive — 12+ hours to read; 180-day minimum; cheapest per-GB.

Seen in

Last updated · 542 distilled / 1,571 read