← Design Cost-Optimized Architectures

▶ Play from here · 71 min

Starts this lesson and continues through 6 more to the end of certification prep.

Storage cost — classes, minimums, lifecycle, and who pays

Why this lesson matters

Task statement 4.1 is "Design cost-optimized storage solutions", and it is the longest task statement in Domain 4: eleven "Knowledge of" items and ten "Skills in" items. The skills that decide most questions, verbatim:

• "Selecting the appropriate storage tier"
• "Managing S3 object lifecycles"
• "Selecting the correct data lifecycle for storage"
• "Selecting the most cost-effective storage service for a workload"
• "Determining the lowest cost method of transferring data for a workload to AWS storage"
• "Designing appropriate storage strategies (for example, batch uploads to Amazon S3 compared with individual uploads)"

And one knowledge item that only has one answer: "Access options (for example, an S3 bucket with Requester Pays object storage)".

The distinguishing question for S3 is: how often is it read, how fast must it come back, and how long will it stay? All three. Leave one out and you walk into a minimum charge.

The S3 storage classes — the table to memorise

From Understanding and managing Amazon S3 storage classes, fetched 2026-09-25. This is AWS's own comparison table, reduced to the columns the exam uses:

Durability is "99.999999999%" (eleven nines) for every class in that table except the not-recommended Reduced Redundancy Storage.

⚠️ 30 / 90 / 180. Standard-IA and One Zone-IA: 30 days. Glacier Instant and Glacier Flexible: 90 days. Deep Archive: 180 days. Standard and Intelligent-Tiering: none.

The minimum charges, in AWS's words

The IA note, verbatim — it defines both minimums at once:

"If an object is less than 128 KB, Amazon S3 charges you for 128 KB. If you delete an object before the end of the 30-day minimum storage duration period, you are charged for 30 days. Objects that are deleted, overwritten, or transitioned to a different storage class before 30 days will incur the normal storage usage charge plus a pro-rated charge for the remainder of the 30-day minimum."

And the Glacier rule, from Understanding S3 Glacier storage classes:

"If you delete, overwrite, or transition the object to a different storage class before the minimum, you are charged for the remainder of that duration."

⚠️ The two classic traps:

1. Short-lived data in IA or Glacier. Temp files kept seven days in Standard-IA are billed for 30. Short-lived data belongs in Standard.
2. Millions of tiny objects in IA. A 4 KB object in Standard-IA is billed as 128 KB — 32 times its size.

The Glacier archival overhead

For Flexible Retrieval and Deep Archive, AWS adds metadata to every object:

"S3 Glacier Flexible Retrieval requires 40 KB of additional metadata for each archived object. This includes 32 KB of metadata charged at the S3 Glacier Flexible Retrieval rate … and an additional 8 KB data charged at the S3 Standard rate."

Same 32 + 8 split for Deep Archive. And AWS's advice follows directly: "If you are archiving small objects, consider these storage charges. Also consider aggregating many small objects into a smaller number of large objects to reduce overhead costs."

That is the answer to the skill "batch uploads to Amazon S3 compared with individual uploads" when it's asked about archive: batch them into fewer, larger objects.

Retrieval times — Glacier Flexible vs Deep Archive

Verbatim from the Glacier page:

Summary row from AWS's table: Instant Retrieval "Milliseconds", Flexible "Minutes to 12 hours", Deep Archive "9 to 48 hours". Deep Archive is "the lowest-cost storage option in AWS".

⚠️ Glacier Instant Retrieval is not archival. It serves in milliseconds, "with the same latency and throughput performance as the S3 Standard-IA storage class", and costs less to store but more to read: "When compared to S3 Standard-IA, S3 Glacier Instant Retrieval has lower storage costs but higher data access costs." So the discriminator between the two is access frequency: once a month → Standard-IA; once a quarter → Glacier Instant Retrieval.

⚠️ Restoring costs double storage for a while. "When you restore an archive, you are paying for both the archive … and a copy that you restored temporarily (S3 Standard storage rate)."

One Zone-IA — when losing an AZ is acceptable

"S3 One Zone-IA … stores the object data in only one Availability Zone, which makes it less expensive than S3 Standard-IA. However, the data is not resilient to the physical loss of the Availability Zone."

AWS's recommendation, verbatim: Standard-IA "for your primary or only copy of data that can't be re-created"; One Zone-IA "if you can re-create the data if the Availability Zone fails, for object replicas when configuring S3 Cross-Region Replication (CRR)".

Exam signal: "thumbnails regenerated from originals", "secondary copy of replicated data" → One Zone-IA. "Only copy", "can't be recreated", "compliance" → never One Zone.

S3 Intelligent-Tiering — when you don't know

From How S3 Intelligent-Tiering works, fetched 2026-09-25:

"For a low monthly object monitoring and automation charge, S3 Intelligent-Tiering monitors access patterns and automatically moves objects to the Infrequent Access tier when they haven't been accessed for 30 consecutive days. After 90 days of no access, the objects are moved to the Archive Instant Access tier without performance impact or operational overhead."

Three facts that decide questions:

• No retrieval fees. "There are no retrieval fees for S3 Intelligent-Tiering." Accessing an object in a lower tier "automatically moved back to the Frequent Access tier".
• No minimum duration and no minimum size in AWS's comparison table.
• ⚠️ Small objects aren't tiered. "If the size of an object is less than 128 KB, it is not monitored and is not eligible for automatic tiering. Smaller objects are always stored in the Frequent Access tier."

Exam signal: "unknown", "unpredictable", "changing" access pattern → Intelligent-Tiering. "Known, predictable" pattern (for example, "accessed for 30 days then rarely") → a lifecycle rule to a specific class, which avoids the per-object monitoring fee.

Lifecycle rules — the waterfall and its constraints

From Transitioning objects using Amazon S3 Lifecycle, fetched 2026-09-25. S3 "supports a waterfall model" — transitions go down, toward colder classes. The supported transitions, from the page:

⚠️ Note the One Zone-IA row: it is not listed as able to transition to Glacier Instant Retrieval.

Two rules that make the direction explicit:

• "Objects that are stored in the S3 Glacier Flexible Retrieval storage class can only be transitioned to the S3 Glacier Deep Archive storage class."
• "The transition of objects to the S3 Glacier Deep Archive storage class can go only one way." To go back up you restore, then copy.

⚠️ The 128 KB default. "Starting September 2024, the default behavior prevents objects smaller than 128 KB from being transitioned to any storage class." The reason is cost: "you are charged a transition request for each object. This means, for smaller objects, the transition costs can outweigh the storage savings." You can override it with an object size filter.

⚠️ You can't chain transitions faster than the minimum. AWS's example: Glacier Instant Retrieval has a 90-day minimum, so a rule that moves objects to Glacier IR after 4 days "the S3 Glacier Deep Archive transition must occur after at least 94 days".

⚠️ Billing starts at the rule date, not the physical move. For transitions to Glacier Flexible or Deep Archive: "You are charged at the destination storage class rate … starting from the date the lifecycle rule is satisfied, even if the physical transition has not yet occurred."

Requester Pays — the only answer to one question

From Using Requester Pays general purpose buckets, fetched 2026-09-25:

"With Requester Pays buckets, the requester instead of the bucket owner pays the cost of the request and the data download from the bucket. The bucket owner always pays the cost of storing data."

Requirements and restrictions, verbatim:

• "If you enable Requester Pays on a general purpose bucket, anonymous access to that bucket is not allowed."
• Requesters must include the x-amz-request-payer header (or --request-payer on the CLI) to show they accept the charges.
• If the requester assumes a role, "the account to which the role belongs is charged".

Exam signal: "share a large dataset with other AWS accounts without paying for their downloads" → Requester Pays. If the stem also says "publicly, to anonymous users" → Requester Pays can't do it.

Data transfer into S3

From Amazon S3 pricing, fetched 2026-09-25: "You pay for all bandwidth into and out of Amazon S3, except for the following" — and the exceptions include "Data transferred in from the internet", "Data transferred between S3 buckets in the same AWS Region", "Data transferred from an Amazon S3 bucket to any AWS service(s) within the same AWS Region", and "Data transferred out to Amazon CloudFront".

Requests are charged per 1,000, by request type (PUT/COPY/POST/LIST, GET/SELECT, lifecycle transition requests, data retrieval requests). ⚠️ The per-request rates on that page are filled in client-side and did not come through in the raw page text, so I don't quote them — read the Requests tab of aws.amazon.com/s3/pricing/ for your Region.

So ingest into S3 is billed mainly by request count, not bytes in — which is why fewer, larger objects are cheaper to upload, transition and archive than many small ones.

Block storage — EBS cost levers

From Amazon EBS volume types and General Purpose SSD volumes, fetched 2026-09-25. The exam item is "Block storage options (for example, hard disk drive [HDD] volume types, solid state drive [SSD] volume types)":

The one fact to carry: gp3 is cheaper than gp2 and decouples performance from size.

"gp3 volumes offer a 20 percent lower price per GiB than General Purpose SSD (gp2) volumes." "It also helps you to scale volume performance independently of volume size."

gp3 includes a baseline of "3,000 IOPS" and "125 MiB/s" in the storage price. gp2 performance "scales with volume size" at "3 IOPS per GiB". So the classic waste pattern is a gp2 volume made oversized just to get IOPS — the skill "Determining the correct storage size for a workload". Migrating is online: "modify the volume type, IOPS, and throughput of your existing volumes without interrupting your Amazon EC2 instances."

st1 is for "Big data, Data warehouses, Log processing"; sc1 for "Throughput-oriented storage for data that is infrequently accessed". Both are "125 GiB - 16 TiB" and neither boots.

Snapshots

From Amazon EBS snapshots:

• "A snapshot is an incremental backup, which means that we save only the blocks on the volume that have changed since the most recent snapshot."
• ⚠️ "Because snapshots are incremental, deleting a snapshot might not reduce your data storage costs." Blocks still referenced by another snapshot are kept.
• "AWS does not automatically back up the data stored on your EBS volumes." Automate with Data Lifecycle Manager or AWS Backup.

From Archive Amazon EBS snapshots: the archive tier "offers up to 75 percent lower snapshot storage costs for snapshots that you plan to store for 90 days or longer". But archiving converts the snapshot: "the incremental snapshot is converted to a full snapshot". AWS's named use cases include "Archiving the only snapshot of a volume, such as end-of-project snapshots".

⚠️ I did not fetch the archive restore time or the archive pricing page. Read docs.aws.amazon.com/ebs/latest/userguide/snapshot-archive-pricing.html and .../restore-archived-snapshot.html before quoting either.

File storage — EFS and FSx cost levers

EFS lifecycle, from Managing storage lifecycle, fetched 2026-09-25:

From EFS pricing: Standard is "SSD-based storage which delivers sub-millisecond latencies"; IA is for data "accessed only a few times a quarter"; Archive for data "accessed a few times a year or less". EFS One Zone is "ideal for applications that do not require the Multi-AZ availability and durability offered by Amazon EFS Regional file systems".

The EFS minimums, from the pricing page's footnotes (verified in the raw page text, 2026-09-25):

"IA and Archive have a minimum billable file size of 128 KiB. Files smaller than 128 KiB can be tiered to these storage classes but will be charged for 128 KiB of storage at the appropriate storage class rate. Archive storage is charged for a minimum storage duration of 90 days, even if the data is deleted before 90 days."

The page also says tiering itself is charged — an "IA Tiering charge applies to data tiered from Standard to IA, and from IA to Standard". ⚠️ The page states no minimum duration for EFS IA; I couldn't retrieve the EFS storage-classes user-guide page to confirm either way (docs.aws.amazon.com/efs/latest/ug/storage-classes.html).

FSx for Windows File Server, from Managing storage:

• HDD vs SSD: HDD "comes at a lower cost relative to SSD storage, but with higher latencies" — for "home directories, user and departmental file shares, and content management systems".
• Data deduplication: "Typical savings average 50–60 percent for general-purpose file shares. Within shares, savings range from 30–50 percent for user documents to 70–80 percent for software development datasets."
• ⚠️ Storage can only grow: "You can only increase a file system's storage capacity; you cannot decrease storage capacity." And "You can't change your file system storage type from SSD to HDD" without a backup-and-restore. Over-provisioning is permanent — size carefully.

Storage type → service (the knowledge item "Storage types with associated characteristics (for example, object, file, block)") is covered in depth in SAA3 lesson 1. For cost, the rule is: pick the cheapest type that meets the access protocol, then the cheapest class within it.

Storage auto scaling — grow, never shrink

The skill "Determining when storage auto scaling is required". From RDS storage autoscaling: it triggers when "Free available space is less than or equal to 10 percent of the allocated storage" and "The low-storage condition lasts at least five minutes". And the cost-relevant limit: "Autoscaling can't decrease the allocated storage."

So storage auto scaling is right for unpredictable growth ("a new mobile gaming application that users are adopting rapidly" is AWS's example). It is not a cost-reduction tool — nothing on AWS's page scales storage back down.

Backup and archival

From What is AWS Backup?: "a fully-managed service that makes it easy to centralize and automate data protection across AWS services". The cost features:

• Incremental backups — "For each successive incremental backup, only the changes … are backed up." ⚠️ "Not all resource types support incremental backups."
• Cold storage tier — "lifecycle policies that automatically transition backups from warm storage to cold storage".

From Backup plan options, the cold-storage rule the exam can test:

"Backups transitioned must be stored in cold storage for a minimum of 90 days, in addition to the time in warm storage. AWS Backup requires the retention to be set for 90 days longer than the 'transition to cold after days' setting."

Cold storage is available only for some resource types — the page lists CloudFormation, DynamoDB with advanced features, EBS (via EBS Snapshot Archive), EFS, SAP HANA on EC2, Timestream, and VMware VMs.

Hybrid storage and getting data in cheaply

The knowledge item is "Hybrid storage options (for example, AWS DataSync, AWS Transfer Family, AWS Storage Gateway)"; the skill is "Determining the lowest cost method of transferring data for a workload to AWS storage".

Remember the free direction: data into S3 from the internet has no transfer charge. Direct Connect inbound is also "$0.00 per GB" (lesson 5).

⚠️ Snowball Edge is not in the HTML guide. The PDF v1.1 lists "AWS Snowball Edge" under 4.2; the HTML guide fetched 2026-09-25 lists only "AWS Outposts". This course follows the HTML and does not teach Snowball. If an offline bulk-transfer question appears, answer it on the principle, not from this module.

Choosing, under exam conditions

Check yourself

1. You store 20 million 8 KB thumbnails in Standard-IA. What are you actually billed for per object?
2. What are the minimum storage durations for Standard-IA, Glacier Instant Retrieval, Glacier Flexible Retrieval and Deep Archive?
3. Why doesn't Intelligent-Tiering help a bucket of 50 KB objects?
4. Who pays for storage in a Requester Pays bucket, and can anonymous users read it?
5. Two reasons to move gp2 to gp3.

Answers

1. 128 KB each — "If an object is less than 128 KB, Amazon S3 charges you for 128 KB." That's 16× the real size. They belong in Standard (or aggregated).
2. 30 / 90 / 90 / 180 days.
3. "If the size of an object is less than 128 KB, it is not monitored and is not eligible for automatic tiering. Smaller objects are always stored in the Frequent Access tier."
4. "The bucket owner always pays the cost of storing data." And no — "anonymous access to that bucket is not allowed."
5. "a 20 percent lower price per GiB", and performance "independently of volume size" — so you stop buying capacity to get IOPS.

Teaching this section

• Draw the 30/90/180 ladder on the board first. Then give three workloads and ask which minimum each one violates. People remember minimums they've been caught by.
• Demo: upload a 4 KB object with --storage-class STANDARD_IA, then show the pricing math. The 32× multiplier is the thing that makes it stick.
• Best discussion prompt: "Intelligent-Tiering has no minimums and no retrieval fees. Why isn't it the default for everything?" Walk them to the per-object monitoring fee and the 128 KB exclusion.
• Watch for: Glacier Instant Retrieval treated as archival. It serves in milliseconds; only Flexible and Deep Archive need a restore.
• Timing: 30 min presented, 20 on the choosing table.