Cloud Storage Tiering
Standard, Infrequent Access, and Glacier — trading cheaper storage for slower, sometimes costlier retrieval.
Overview
Cloud object storage prices data by how often it's accessed — Standard for frequently-accessed data, Infrequent Access for occasional reads, and deep archive tiers like Glacier for data rarely touched — with cheaper storage cost traded against higher retrieval cost and latency at each colder tier.
Why It Exists
Not all stored data is accessed equally often, but a single flat storage price doesn't reflect that — most data, like backups, logs, and old records, is written once and rarely, if ever, read again. Tiering exists to let a provider, and a customer, pay dramatically less for storage that's rarely accessed, in exchange for accepting slower and sometimes costlier retrieval when it actually is needed.
Real World Example
A company's compliance logs are written daily but almost never read. Moving them from S3 Standard to Glacier Deep Archive after 90 days, via an automated lifecycle policy, can cut storage cost by well over 90% — at the cost of retrieval taking hours instead of milliseconds, on the rare occasion they're actually needed for an audit.
Example Data
The tiering trade-off, roughly — exact pricing varies by provider and region
| Tier | Relative Storage Cost | Retrieval Time |
|---|---|---|
| Standard | Highest | Milliseconds |
| Infrequent Access | ~50% less | Milliseconds, plus a per-GB retrieval fee |
| Glacier / Deep Archive | ~80–95% less | Minutes to hours |
The Tiers, One at a Time
Standard — Frequent Access, Instant Retrieval
The default tier: highest storage cost, but data is available with millisecond latency and no retrieval fee — the right choice for anything read regularly.
Infrequent Access — Cheaper Storage, Retrieval Fee
Lower storage cost in exchange for a per-request or per-GB retrieval fee. Still available quickly, but only cost-effective for data that's genuinely read rarely.
Glacier / Deep Archive — Cheapest Storage, Hours-Long Retrieval
The lowest possible storage cost, but retrieval can take anywhere from minutes to hours before the data is even readable — appropriate only for data that truly won't be needed on short notice.
Lifecycle Policies — Automating the Move Between Tiers
Rules that automatically transition objects to colder tiers after a defined age, so tiering happens without manual intervention as data naturally ages out of frequent use.
Diagram
A lifecycle policy moves data to colder tiers automatically as it ages
Object written
Standard tier
After N days (lifecycle rule)
moved to Infrequent Access
After more days (lifecycle rule)
moved to Glacier / Deep Archive
Common Mistakes
Moving small, frequently-accessed files to a colder tier purely to save on storage cost
Why: Colder tiers often charge a per-request or per-GB retrieval fee — for data accessed more often than expected, those fees can exceed the storage savings entirely, making the move a net loss.
Fix: Base tiering decisions on real, measured access patterns, not an assumption that 'old' automatically means 'never accessed.'
Archiving very small, numerous objects into a deep archive tier
Why: Some archive tiers apply a minimum billable object size or storage duration, so archiving millions of tiny files can cost far more than expected, or deliver much smaller savings than the sticker-price storage-cost difference suggests — the small-file cost trap.
Fix: Check the specific tier's minimum size and duration billing rules before archiving very small or very numerous objects.
Assuming retrieval from the coldest tiers is fast enough for anything time-sensitive
Why: Deep archive tiers can take hours to restore data before it's even readable — treating this like Standard's millisecond access for something with a real deadline causes real operational surprises.
Fix: Only move data to deep archive tiers when confident it won't be needed on short notice, and plan retrieval time into any process that might need it back.
Interview Questions
What's the basic trade-off cloud storage tiering offers?
Colder tiers offer significantly cheaper storage in exchange for slower retrieval, and often a retrieval fee — the right tier depends on how frequently the data actually needs to be read.
Why might moving data to a colder storage tier actually cost more money overall in some cases?
If the data ends up being accessed more often than assumed, the per-request or per-GB retrieval fees charged by colder tiers can add up to more than the storage savings gained by moving it there — the tier choice should be based on real measured access frequency, not a guess.
Design a lifecycle policy for a system ingesting 10TB/day of logs, needing fast access to the last 30 days for debugging, occasional access to the last year for compliance, and 7-year retention for legal reasons. What tiers and transitions would you use?
Keep the first 30 days in Standard for fast debugging access. Transition to Infrequent Access after 30 days, covering the up-to-a-year compliance-access window with acceptable retrieval latency and fees for occasional reads. Transition to Glacier or Deep Archive after 1 year, since data that old is very unlikely to be needed quickly, keeping it at minimal cost until the 7-year retention period expires and it can be deleted entirely via the same lifecycle policy.
Production Best Practices
Do
✓Base tiering decisions on real, measured access frequency.
✓Check minimum size/duration billing rules before archiving small or numerous objects.
✓Automate tier transitions with lifecycle policies rather than manual moves.
Don't
✗Don't move data to a colder tier without checking its actual access frequency first.
✗Don't archive large numbers of very small objects without checking the tier's billing minimums.
✗Don't assume deep-archive retrieval is fast enough for time-sensitive needs.
Comparison
| Storage Cost | Retrieval Time | Retrieval Cost | |
|---|---|---|---|
| Standard | Highest | Milliseconds | None |
| Infrequent Access | Lower | Milliseconds | Per-GB retrieval fee |
| Glacier / Deep Archive | Lowest | Minutes to hours | Retrieval fee, sometimes expedite fee |