Cargando...
Cargando...
Choose the right S3 storage class every time — cost, latency, retrieval, and exam traps decoded
Access frequency and retrieval tolerance determine your optimal storage class
| Feature | S3 Standard Frequent access, zero retrieval fees | IA Lower storage cost, pay-per-retrieval | Glacier Archival storage, minutes-to-hours retrieval | Intelligent-Tiering Auto-optimizes cost for unknown patterns |
|---|---|---|---|---|
Designed For The keyword 'infrequently accessed' in a scenario = strong signal for S3-IA or Glacier. 'Unknown access pattern' = Intelligent-Tiering. | Frequently accessed data (>1x/month) | Infrequently accessed data (<1x/month), rapid access when needed | Long-term archival, rarely accessed (1–2x/year) | Unknown or unpredictable access patterns |
Availability SLA S3 Standard-IA has a LOWER availability SLA (99%) than S3 Standard (99.9%). This is a classic exam trap — lower availability means it's not suitable for mission-critical, always-on workloads. | 99.99% availability (99.9% SLA) | 99.9% availability (99% SLA) | 99.99% availability (no retrieval SLA for standard retrieval) | 99.9% availability (99% SLA for IA tier; 99.99% for frequent tier) |
Durability ALL S3 storage classes share the same 11-nines durability. Durability is NEVER the differentiator between classes on the exam — availability and cost are. | 99.999999999% (11 nines) across ≥3 AZs | 99.999999999% (11 nines) across ≥3 AZs | 99.999999999% (11 nines) across ≥3 AZs | 99.999999999% (11 nines) across ≥3 AZs |
Minimum Storage Duration Charge Critical cost trap: if you store an object for 10 days in S3-IA and delete it, you are STILL charged for 30 days. Glacier Flexible Retrieval charges 90 days minimum. Glacier Deep Archive charges 180 days minimum. | None — no minimum duration | 30 days minimum — deleted/transitioned objects before 30 days still billed for 30 days | 90 days minimum (Flexible Retrieval) — objects deleted before 90 days still billed for 90 days | 30 days minimum monitoring charge applies |
Minimum Object Size Charge Storing millions of tiny objects (<128 KB) in S3-IA is MORE expensive than S3 Standard. Intelligent-Tiering automatically keeps small objects in the Frequent Access tier — no retrieval fee, no monitoring overhead penalty for small objects. | No minimum object size | 128 KB minimum — objects smaller than 128 KB are billed AS IF they are 128 KB | 40 KB minimum (8 KB metadata + 32 KB index overhead billed per object) | No minimum object size, but objects <128 KB are never moved to lower tiers (always billed at Frequent Access rates) |
First-Byte Latency S3-IA has the SAME millisecond latency as S3 Standard — the 'infrequent' refers to how often you ACCESS it for cost optimization, NOT a technical access delay. Glacier introduces real retrieval latency. | Milliseconds (immediate) | Milliseconds (immediate, same as Standard) | Minutes to hours depending on retrieval tier (Expedited: 1–5 min, Standard: 3–5 hrs, Bulk: 5–12 hrs) | Milliseconds for Frequent/Infrequent tiers; Archive tiers require async retrieval initiation |
Retrieval Fee S3 Standard is the ONLY traditional class with zero retrieval fees. If a workload reads data frequently, retrieval fees in S3-IA can quickly exceed the storage savings. Intelligent-Tiering eliminates retrieval fees for the two automatic tiers. | None — no per-GB retrieval charge | Per-GB retrieval fee applies on every GET | Per-GB retrieval fee applies; varies by retrieval speed tier (Expedited > Standard > Bulk pricing) | No retrieval fees for Frequent or Infrequent Access tiers; Archive tiers have retrieval fees |
Storage Cost (Relative) Prices are region-specific and change over time — memorize relative cost order for exams: Standard > IA > Glacier Flexible > Glacier Deep Archive. The monitoring fee for Intelligent-Tiering makes it cost-ineffective for small numbers of objects. | Highest per-GB storage price (~$0.023/GB/month in us-east-1) | ~40% cheaper than Standard for storage (~$0.0125/GB/month in us-east-1) | ~80–90% cheaper than Standard (~$0.004/GB/month Flexible Retrieval in us-east-1) | Frequent tier = Standard pricing; IA tier = IA pricing; monitoring fee ~$0.0025 per 1,000 objects/month |
Intelligent-Tiering Automatic Tiers Archive Access and Deep Archive Access tiers in Intelligent-Tiering are OPTIONAL and must be explicitly activated. The first three tiers (Frequent, Infrequent, Archive Instant) are automatic with no retrieval fees. | N/A — single tier | N/A — single tier | N/A — single tier (manual lifecycle or direct PUT) | Frequent Access (default) → Infrequent Access (30 days no access) → Archive Instant Access (90 days) → Archive Access (90–180+ days, optional) → Deep Archive Access (180+ days, optional) |
Lifecycle Policy Support Lifecycle transitions are ONE-WAY DOWNWARD — you cannot use a lifecycle policy to move objects from Glacier back to Standard. To restore, you must use a Glacier restore operation which creates a temporary Standard copy. | Source class — can transition to IA, Glacier, or Deep Archive | Can transition to Glacier or Deep Archive; can be transitioned FROM Standard | Can transition to Glacier Deep Archive; cannot transition BACK to Standard or IA via lifecycle | Can be a lifecycle target; internally manages its own tiering automatically |
Glacier Retrieval Options Expedited retrievals from Glacier can be provisioned with 'Provisioned Capacity Units' to guarantee availability of fast retrievals. Without provisioned capacity, Expedited retrieval may fail during high-demand periods. | N/A | N/A — immediate retrieval, no options needed | Expedited (1–5 min, highest cost), Standard (3–5 hrs, moderate cost), Bulk (5–12 hrs, lowest cost) | Archive Instant: milliseconds; Archive/Deep Archive tiers: same as Glacier retrieval options |
Use Case Examples | Static website assets, active application data, frequently accessed logs, media serving, ML training datasets in use | Disaster recovery backups, secondary copies of on-prem data, older log files still needed for compliance, monthly reports | Long-term regulatory archives (7+ years), media master files, raw scientific data, audit logs, tape replacement | Data lakes with mixed access patterns, user-generated content, new datasets with unknown future access, large content repositories |
S3 Object Lock / WORM Support For compliance/WORM requirements, all S3 storage classes support Object Lock. Glacier Vault Lock (legacy API) is separate from S3 Object Lock but serves the same WORM purpose. | Yes — Compliance and Governance modes supported | Yes — Compliance and Governance modes supported | Yes — Compliance and Governance modes supported (Vault Lock in legacy Glacier API) | Yes — Object Lock applies at the bucket level, compatible with all tiers |
Cross-Region Replication (CRR) Support You can configure CRR to replicate Standard objects to an IA or Glacier class in the destination bucket to save costs on replicated copies. | Yes — fully supported as source and destination | Yes — can replicate to same or different storage class in destination | Yes — S3 Glacier Flexible Retrieval can be a replication destination; objects replicated to Glacier are immediately accessible per Glacier retrieval rules | Yes — fully supported; replication respects tiering in destination bucket |
S3 Select / Glacier Select Glacier Select allows SQL queries directly against archived data, avoiding a full restore. This is a cost-optimization pattern for analytics on cold data. | S3 Select supported — query CSV/JSON/Parquet in-place | S3 Select supported | Glacier Select supported — query archived data without full restore (SQL-based, charged per query) | S3 Select supported for accessible tiers |
Encryption at Rest S3 Glacier encrypts data by default using AES-256 (SSE-S3). You don't need to configure encryption — it's automatic. This is different from S3 Standard where default encryption must be configured (though it is now on by default for new buckets). | SSE-S3, SSE-KMS, SSE-C, DSSE-KMS all supported | SSE-S3, SSE-KMS, SSE-C, DSSE-KMS all supported | SSE-S3 by default (AES-256); SSE-KMS supported | SSE-S3, SSE-KMS, SSE-C, DSSE-KMS all supported |
Monitoring / Automation Fee The Intelligent-Tiering monitoring fee is small per object but adds up at scale for tiny objects. AWS recommends Intelligent-Tiering only for objects ≥128 KB with unknown access patterns. | None | None | None | Per-object monitoring fee (~$0.0025 per 1,000 objects/month) — makes it cost-ineffective for large numbers of tiny, cheap objects |
Summary
S3 Standard is the default for any data accessed frequently or requiring immediate, low-latency access without retrieval cost concerns. S3 Standard-IA is ideal when data is accessed less than once a month, objects are large enough (>128 KB) and stored long enough (>30 days) to justify the lower storage price despite per-retrieval fees. S3 Glacier is purpose-built for long-term archival where retrieval latency of minutes to hours is acceptable and storage cost minimization is paramount. S3 Intelligent-Tiering is the 'set-and-forget' choice when access patterns are genuinely unknown or highly variable, automatically moving objects between tiers to optimize cost without operational overhead.
🎯 Decision Tree
If data is accessed frequently (daily/weekly) → S3 Standard. If data is rarely accessed (<1x/month), objects >128 KB, stored >30 days, but needs millisecond retrieval when accessed → S3 Standard-IA. If data is archival (accessed 1–2x/year), retrieval latency of hours is acceptable, and cost minimization is the priority → S3 Glacier Flexible Retrieval. If data is never expected to be retrieved (>7 year compliance hold) → S3 Glacier Deep Archive. If access patterns are UNKNOWN or unpredictable, objects are >128 KB → S3 Intelligent-Tiering. If you need WORM/compliance archival with audit trail → S3 Glacier with Vault Lock OR S3 with Object Lock in Compliance mode.
CRITICAL — Minimum Duration Charges Are Cost Traps: S3 Standard-IA charges a 30-day minimum, Glacier Flexible Retrieval charges 90 days, Glacier Deep Archive charges 180 days. If a scenario involves short-lived objects, frequent overwrites, or objects deleted quickly, S3 Standard is cheaper despite higher per-GB storage cost. Always calculate total cost including minimum duration penalties.
CRITICAL — S3-IA Availability SLA is 99%, NOT 99.9%: S3 Standard-IA has a lower availability SLA than S3 Standard. Exam scenarios asking for 'highly available' or 'mission-critical' storage should use S3 Standard. S3-IA is explicitly for data that can tolerate slightly lower availability in exchange for cost savings.
CRITICAL — S3-IA Retrieval Latency is Milliseconds (Same as Standard): The word 'infrequent' in S3 Standard-IA refers to HOW OFTEN you access data for cost-optimization purposes, NOT a technical delay. S3-IA delivers data in milliseconds just like S3 Standard. Only Glacier introduces actual retrieval latency. Confusing this is the #1 trap for CLF-C02 and SAA-C03 candidates.
IMPORTANT — Intelligent-Tiering is NOT Free Optimization: There is a per-object monitoring fee (~$0.0025/1,000 objects/month). For workloads with millions of tiny objects or KNOWN access patterns, Intelligent-Tiering adds cost without benefit. Use explicit storage classes when access patterns are predictable.
IMPORTANT — 128 KB Minimum Object Size in S3-IA: Objects smaller than 128 KB in S3 Standard-IA are billed as 128 KB. Storing many small objects (thumbnails, metadata files, JSON configs) in S3-IA is an anti-pattern that costs MORE than S3 Standard. In Intelligent-Tiering, objects <128 KB never leave the Frequent Access tier.
IMPORTANT — Lifecycle Transitions Are One-Way (Downward Only): You cannot use S3 Lifecycle policies to move objects from Glacier back to Standard or IA. Restoration from Glacier creates a temporary copy in S3 Standard that expires after a specified period (1–365 days). The original archived object remains in Glacier.
IMPORTANT — Glacier Expedited Retrieval Requires Provisioned Capacity for Guarantees: Without Provisioned Capacity Units, Expedited retrievals (1–5 min) can be rejected during high demand. For scenarios requiring guaranteed fast access to Glacier data, mention Provisioned Capacity. Each unit provides at least 3 expedited retrievals per 5 minutes and up to 150 MB/s of retrieval throughput.
NICE-TO-KNOW — All S3 Storage Classes Share 11-Nines Durability: Durability (99.999999999%) is NEVER the differentiator between S3 storage classes. Every exam question comparing storage classes on durability has the same answer for all classes. The differentiators are: availability, latency, cost, retrieval fees, and minimum duration.
NICE-TO-KNOW — Glacier Select Enables SQL Queries on Archived Data: You can run SQL queries directly on data stored in S3 Glacier without performing a full restore. This is useful for compliance audits or analytics on cold data where only a subset of records is needed. Charged per amount of data scanned and returned.
NICE-TO-KNOW — S3 Glacier Instant Retrieval vs Flexible Retrieval: AWS has two separate Glacier tiers that confuse many candidates. Glacier Instant Retrieval = millisecond access, 90-day minimum, higher cost than Flexible. Glacier Flexible Retrieval = minutes-to-hours access, 90-day minimum, lower cost. Glacier Deep Archive = 12–48 hour retrieval, 180-day minimum, cheapest of all.
The #1 exam trap: Candidates assume S3 Standard-IA has slow retrieval because of the word 'Infrequent Access' — it does NOT. S3-IA delivers data in milliseconds, identical to S3 Standard. The 'infrequent' refers to cost optimization assumptions (you pay a retrieval fee each time), not a technical access delay. Only Glacier (Flexible Retrieval and Deep Archive) introduces actual retrieval latency measured in minutes to hours. A close second trap: assuming all S3 classes have the same availability SLA — S3 Standard-IA has a 99% SLA vs 99.9% for S3 Standard.
CertAI Tutor · SAA-C03, SAP-C02, DOP-C02, SCS-C02, AIF-C01, CLF-C02, DVA-C02, DEA-C01 · 2026-02-22
Services
Comparisons