← Study home

🪣 Amazon S3 & Storage Classes — Deep Dive

Proper understanding of S3 and all its storage classes for AWS-Cloud-Practitioner (Domain 3, Module 6). Companion to the quick kitchen-analogy version in the hub — this is the "actually understand why" reference.

← back to AWS-Cloud-Practitioner · Networking · DevOps

What S3 actually is

Object storage: store objects (data + metadata + a unique key) in buckets, reached by URL/API (not mounted as a disk). - Stored in a Region; most classes auto-replicate across 3+ AZs. - 11 nines durability (99.999999999%) — effectively never loses data. - Unlimited total; single object up to 5 TB. Bucket names are globally unique. - Uses: backups, static websites, data lakes, media, app/static assets.

The one model — a single cost dial with 3 levers

Every class is a trade on: ① storage cost (to keep it) · ② retrieval cost + speed (to read it) · ③ availability / AZ count.

The deal: the less often you read it and the longer you'll wait to get it back, the cheaper AWS stores it — but the more you pay (money + time) to retrieve it.

So you only ask two questions: How often will I read this? and How fast do I need it when I do?

Key truth: all classes share the same 11-nines durability — they are NOT safer-vs-riskier. They differ on cost, retrieval speed, and availability.

The classes, hot → cold

Class	Storage $	Read $ / speed	AZs	Min keep	Why it exists
S3 Standard	Highest	Free, instant	3+	none	Active data touched constantly; free instant reads
S3 Intelligent-Tiering	Auto	No retrieval fees, instant	3+	none	Unknown/changing pattern — AWS monitors & moves each object for a tiny fee
S3 Standard-IA	Lower	Pay per GB, instant	3+	30 days	Known-infrequent but needed instantly
S3 One Zone-IA	~20% under Std-IA	Pay per GB, instant	1	30 days	Same as Std-IA but single AZ — cheaper, lost if AZ destroyed → recreatable data only
S3 Glacier Instant Retrieval	Cheap	Pay, instant (ms)	3+	90 days	Archive touched ~quarterly but needed in milliseconds
S3 Glacier Flexible Retrieval	Cheaper	Pay, minutes–hours	3+	90 days	Archive where you can wait (the classic "Glacier")
S3 Glacier Deep Archive	Cheapest	Pay, 12–48 hrs	3+	180 days	Long-term/compliance ("keep 7 yrs"); almost never read

Going down: storage cheaper · retrieval pricier + slower · minimum storage period longer (so never put fast-changing data in IA/Glacier).

The three confusions, resolved

Intelligent-Tiering vs Standard-IA — both = infrequent access. Std-IA = you know it's infrequent (accept a retrieval fee). Intelligent-Tiering = you don't know → AWS decides, no retrieval fees, small monitoring fee. Trigger: "unknown/changing" → Intelligent-Tiering; "known infrequent" → Standard-IA.
One Zone-IA vs Standard-IA — identical except AZs. Std-IA = 3+ AZs (survives AZ loss); One Zone-IA = 1 AZ (cheaper, lost if AZ dies). Trigger: "can tolerate loss / recreatable / non-critical" → One Zone-IA.
The three Glaciers — all archival; differ only by retrieval speed: Instant (ms) · Flexible (min–hrs) · Deep Archive (hrs, cheapest). The question's retrieval-time tolerance picks which.

Durability vs Availability (don't conflate)

Durability = "will my data survive?" → all S3 = 11 nines (equally safe; One Zone's caveat is only a whole-AZ destruction).
Availability = "can I reach it right now?" → drops as you go colder (Standard ~99.99% → One Zone ~99.5% → Glacier needs a restore). Cheaper classes trade away availability, not durability.

Lifecycle policies

Rules that slide objects down the dial as they age — e.g. Standard (0–30d) → Standard-IA (30–90) → Glacier Flexible (90–365) → Deep Archive / delete.

Exam cue: "automatically transition older data to cheaper storage" → lifecycle policy.

Exam trigger words → class

Question says...	Class
"frequently accessed" / default	S3 Standard
"unknown / changing / unpredictable" + "automatic" cost optimization	S3 Intelligent-Tiering
"infrequently accessed" but "available immediately / rapid"	S3 Standard-IA
infrequent + "can tolerate" single-AZ loss / "reproducible"	S3 One Zone-IA
"archive / long-term / rarely" + retrieval-time tolerance	S3 Glacier (ms→Instant · min–hrs→Flexible · hrs + cheapest + compliance→Deep Archive)

Worked examples

Profile images read thousands of times/day → Standard.
Patient scans, rarely opened but must load instantly → Standard-IA.
Startup, no idea which files are hot → Intelligent-Tiering.
Financial records, 7 years, almost never read, hours OK → Glacier Deep Archive.
Backups accessed ~quarterly but needed instantly → Glacier Instant Retrieval.
Easily-recreated thumbnails, infrequent, cost-sensitive → One Zone-IA.

Source

Taught in the AWS CLF-C02 study session, 2026-06-30 (Module 6 Storage), when the quick analogy wasn't enough — Sahil wanted to understand the why behind every class. Mechanics-level companion to the hub's exam-level Storage section.