Choosing an S3 Storage Class for Infrequently Accessed Data: Decision Guide

Decision and constraints

You need to store objects that are accessed less than once per month, require 99.9 % durability, sub‑second retrieval latency, and you want to automate tiering with lifecycle policies while keeping monthly storage cost low.

Supported options

Storage class	Storage cost (USD/GB‑month)	GET request cost (USD/1000)	Typical retrieval latency	Minimum storage duration	Typical use‑case
S3 Standard‑IA	≈ 0.0125	≈ 0.01	milliseconds (first‑byte)	30 days	Data needing multi‑AZ resilience
S3 One Zone‑IA	≈ 0.01	≈ 0.01	milliseconds	30 days	Reproducible or backup‑eligible data
S3 Glacier Instant Retrieval	≈ 0.004	≈ 0.025	sub‑second (typically < 500 ms)	90 days	Archive‑like access with fast retrieve

Trade‑offs

• Standard‑IA provides redundancy across at least two Availability Zones, protecting against an AZ‑wide outage, but costs ~20 % more than One Zone‑IA.
• One Zone‑IA stores all copies in a single AZ, lowering cost; if that AZ suffers an outage the data is unavailable unless you have another copy elsewhere.
• Glacier Instant Retrieval offers the lowest storage price, but GET request fees are higher and a 90‑day minimum storage duration applies; frequent retrieval can erase the storage savings.

Implementation: lifecycle rule JSON

The following rule moves objects to STANDARD_IA after 30 days, then to GLACIER_IR after an additional 60 days (total 90 days). Adjust the prefixes or tags to match your data set.

{
  "Rules": [
    {
      "ID": "Move-to-IA-then-GlacierIR",
      "Status": "Enabled",
      "Filter": { "Prefix": "infrequent/" },
      "Transitions": [
        { "Days": 30, "StorageClass": "STANDARD_IA" },
        { "Days": 90, "StorageClass": "GLACIER_IR" }
      ],
      "NoncurrentVersionTransitions": [],
      "Expiration": { "Days": 3650 }
    }
  ]
}

To apply the rule, run (requires s3:PutLifecycleConfiguration permission on the bucket):

aws s3api put-bucket-lifecycle-configuration \
    --bucket <bucket-name> \
    --lifecycle-configuration file://lifecycle.json

Validation steps

1. Upload a test object with the prefix infrequent/ to the bucket.
2. Wait >30 days (or adjust the rule for testing with a shorter period).
3. Run aws s3api head-object --bucket <bucket-name> --key infrequent/test-object and verify that StorageClass equals STANDARD_IA.
4. After >90 days total, repeat the head-object command; the response should show StorageClass: GLACIER_IR.
5. Optionally, time a GET request: time aws s3 cp s3://<bucket-name>/infrequent/test-object /dev/null. The elapsed time should be under 500 ms for GLACIER_IR.
6. Enable S3 Inventory for the bucket (requires s3:PutInventoryConfiguration) and check the weekly CSV for rows where the storage class transitions as expected.

Limitations and checks

• One Zone‑IA does not survive an AZ‑wide failure; ensure the data can be regenerated or is mirrored elsewhere.
• Glacier Instant Retrieval charges higher per‑GET fees; estimate your request volume before selecting this class.
• Lifecycle transitions only take effect at midnight UTC after the day count is met; allow for this delay when validating.