Storage costs

Key points

• Minimum duration: 30 days for Standard-IA and One Zone-IA, 90 days for Glacier Instant Retrieval and Glacier Flexible Retrieval, 180 days for Glacier Deep Archive. Deleting earlier is billed as if kept.
• Minimum billable size: 128 KB for Standard-IA, One Zone-IA and Glacier Instant Retrieval. Glacier Flexible Retrieval and Deep Archive add 40 KB of metadata per object.
• Retrieval fees: per GB for the IA and Glacier classes. Standard and Intelligent-Tiering have none.
• One Zone-IA: one AZ and cheaper than Standard-IA, for data you can re-create, such as replicas or derived files.
• Express One Zone: single-digit millisecond access in one AZ with request costs 50% lower than Standard. A performance class, not an archive.

Key points

• Transition actions: move objects by age, prefix, tag or size, from Standard down to IA, Intelligent-Tiering, Glacier and Deep Archive. Deep Archive is a one-way door.
• Expiration actions: delete current versions, permanently delete noncurrent versions with NoncurrentVersionExpiration and clean up expired delete markers.
• Small objects: by default objects under 128 KB are not transitioned, because the per-object transition request can cost more than it saves. A size filter overrides it.
• Minimum duration: moving out of a class early bills the rest of its minimum. One rule cannot go to Glacier Instant Retrieval at day 4 and Deep Archive at day 20, the second step must wait until day 94.
• Billing date: transitions to Glacier Flexible Retrieval and Deep Archive bill at the new rate from the day the rule is met. Intelligent-Tiering bills after the physical move.

Key points

• Automatic tiers: Frequent Access, Infrequent Access after 30 days without access, Archive Instant Access after 90 days. Any access moves the object back to Frequent Access.
• Optional archive tiers: Archive Access from 90 days and Deep Archive Access from 180 days, configurable up to 730 days. Objects there must be restored before reading.
• Costs: a small monthly monitoring and automation fee per object, no retrieval fees and no minimum storage duration.
• Small objects: under 128 KB they are not monitored and always stay in Frequent Access.
• Adoption: set it as the storage class at upload, or move existing objects into it with a Lifecycle rule.

Key points

• S3 Storage Lens: organization-wide view down to account, Region, storage class, bucket and prefix. Flags buckets without lifecycle rules for incomplete multipart uploads older than 7 days. Free metrics plus a paid advanced tier.
• Storage Class Analysis: observes a bucket, prefix or tag for 30 days or more and tells when to move Standard objects to Standard-IA. It recommends nothing else.
• Requester Pays: the requester pays for requests and data download, the bucket owner still pays storage. Requests must be authenticated and send x-amz-request-payer, anonymous access is not allowed.

Key points

• gp2 to gp3: gp3 costs 20% less per GiB, includes 3,000 IOPS and 125 MiB/s at any size, and scales IOPS and throughput apart from size. Change it in place with Elastic Volumes, no downtime.
• gp2 trap: IOPS grow at 3 per GiB, so teams buy capacity just to get IOPS. gp3 removes that.
• HDD types: st1 for frequent, large sequential throughput (EMR, ETL, logs), sc1 the lowest cost for cold sequential data. Neither can be a boot volume.
• Snapshots Archive: up to 75% lower snapshot storage cost for snapshots kept 90 days or more. Stored as full snapshots, 90-day minimum, restore takes up to 72 hours.
• Clean up: unattached volumes and old snapshots keep billing. Compute Optimizer and Trusted Advisor flag underutilized volumes.

Key points

• EFS classes: Standard (sub-millisecond), Infrequent Access (up to 95% cheaper) and Archive (up to 50% cheaper than IA, 90-day minimum). IA and Archive bill at least 128 KiB per file.
• EFS lifecycle management: by default moves files not accessed for 30 days to IA and for 90 days to Archive. It can move a file back to Standard on first access.
• EFS One Zone: data in a single AZ, for data that does not need multi-AZ resilience. Access from another AZ pays $0.01 per GB cross-AZ transfer. Archive needs a Regional file system with Elastic Throughput.
• FSx for Windows File Server: HDD for home directories and shares, SSD for latency-sensitive work. Data Deduplication typically saves 50 to 60% on general file shares.
• FSx for Lustre: scratch file systems for short, temporary processing with no replication, persistent ones for long-term data. The Intelligent-Tiering storage class moves cold data to cheaper tiers.

Key points

• Glacier Instant Retrieval: millisecond reads with no restore step, for archives read about once a quarter.
• Bulk is free: Bulk retrievals cost nothing from Glacier Flexible Retrieval and Intelligent-Tiering. Expedited is a premium option.
• Provisioned capacity: guarantees Expedited retrievals from Glacier Flexible Retrieval when demand is high.
• Restored copy: a temporary copy is billed at the Standard rate while the archived object keeps billing at its own rate.
• Large restores: up to 1 to 2 PB per day per account, and S3 Batch Operations to restore many objects with one request.

Key points

• AWS DataSync: managed online transfer from NFS, SMB, HDFS, object storage and other clouds to S3, EFS and FSx, or straight to Glacier classes. Incremental, compressed, validated and encrypted, and one task can fill 10 Gbps.
• Slow links: for a large migration, run DataSync over a hosted Direct Connect connection bought only for the project.
• Snowball Edge: no longer available to new customers. AWS points them to DataSync, to AWS Data Transfer Terminal for physical uploads and to Outposts for edge compute.
• AWS Transfer Family: managed SFTP, FTPS, FTP, AS2 and browser-based transfers into S3 or EFS, so partners keep their clients and you run no servers.
• Batch uploads: fewer, larger objects cut per-request charges and lifecycle transition costs.