Skip to main content
INS // Insights

EFS Infrequent Access Retrieval Cost

Updated November 2026 · 8 min read

Amazon EFS Infrequent Access retrieval cost is what a read does to files that lifecycle management already moved out of Standard. The storage rate drops. The read is a different line. A backup, a malware scan, a tar, an rsync, or a training job that opens every file pays that line for the bytes it reads. A directory listing does not. Teams turn lifecycle on, watch the storage line fall, and meet the retrieval line a month later when something reads the tree.

Rutagon reads this meter before anyone "optimizes" the lifecycle policy. AWS cloud infrastructure work includes the file system and the bill beside it. The scorecard separates storage, tiering, Infrequent Access data access, and throughput. Those are not one number. S3 lifecycle transition cost is the object-storage cousin. EFS is a file system: metadata stays in Standard, and a POSIX read of file bytes is the retrieval.

Where the Amazon EFS Infrequent Access retrieval cost is metered

AWS documents lifecycle behavior in Managing storage lifecycle. Three policies exist. Transition into Infrequent Access moves files that have not been accessed in Standard. The default window is 30 days. Transition into Archive moves colder files. The default window is 90 days without access in Standard. Transition into Standard decides whether a later read moves the file back. The default is that files stay in Infrequent Access or Archive when they are read.

The access timer is internal. It is not the POSIX timestamp you can stat from a client. Listing a directory does not count as file access. Metadata, including names, ownership, and the directory tree, stays in Standard so metadata operations keep Standard latency. During a transition, the file content is still billed as Standard. A write to a file that already sits in Infrequent Access or Archive is written to Standard first and becomes eligible to transition again after 24 hours.

The billing names are in Understanding billing and usage reports for Amazon EFS. Infrequent Access reads and tiering show up as IADataAccess-Bytes. Elastic Throughput reads and writes show up as ETDataAccess-Bytes. Storage is IATimedStorage usage, including a small-file line. CloudWatch StorageBytes for the Infrequent Access class tells you how many bytes are sitting there. It does not tell you how many bytes were read. A dashboard that only graphs storage will celebrate the lifecycle policy while retrieval climbs on the bill.

The EFS FAQ walks a US East (N. Virginia) example. In that example, Infrequent Access read access is priced at $0.01 per GB, and Elastic Throughput reads of those same bytes are priced at $0.03 per GB. Tiering from Standard into Infrequent Access is $0.01 per GB in the same example. Standard storage in the example is $0.30 per GB-month, and the example's Infrequent Access storage math uses $0.0165 per GB-month. A feature page has published a nearby headline for Regional Elastic Infrequent Access storage. When two pages disagree by a fraction of a cent, we use the Cost Explorer line for that Region, not a blended headline.

region: us-east-1 example from the EFS FAQ
ia_read_access: $0.01 per GB
elastic_throughput_read: $0.03 per GB
ia_storage_in_that_example: $0.0165 per GB-month
standard_storage_in_that_example: $0.30 per GB-month
listing_a_directory: not file access, not IADataAccess-Bytes
storage_bytes_metric: size sitting in IA, not bytes read

One full read of a gigabyte, in that illustration, is $0.04 if the file system is on Elastic Throughput: one cent of Infrequent Access access plus three cents of throughput. Storing that gigabyte in Infrequent Access for a month is about a cent and a half in the same illustration. A single full pass costs more than a month of the cheap storage. A daily full pass is on the order of thirty times $0.04, which is well above the $0.30 Standard storage price. A weekly full pass is four reads, about $0.16 of retrieval, plus the Infrequent Access storage that never went away. The "much cheaper than Standard" headline survives only when reads are rare.

Those dollars are the FAQ's N. Virginia illustration. Other Regions publish other rates. Provisioned Throughput does not add the Elastic per-GB read line. It bills the throughput you provisioned. The Infrequent Access access charge still applies. We read the throughput mode before we add $0.03. ETDataAccess-Bytes on the usage report is the sign the elastic line is real. ProvisionedTP-MiBpsHrs is the sign it is not.

Readers that make the line move

The retrieval event is a read of file bytes, not a walk of names. A job that stats the tree to measure it will not move IADataAccess-Bytes. A job that opens each file will. Malware scanners, content indexing, home-grown backups that tar the mount, and data scientists who point a loader at the dataset are the usual readers. We do not assume AWS Backup is innocent or guilty. EFS backup storage and restore are their own charges. We look at whether IADataAccess-Bytes moves during the backup window. If it moves, that backup is a reader. If it does not, the reader is some other client on the mount.

Replication reads the source. AWS charges the prevailing read and tiering rates for source and destination. A replication configuration aimed at a file system that is mostly Infrequent Access is another full-scale reader. Cross-AZ client access is yet another line: the pricing page charges the EC2 cross-AZ rate, stated there as $0.01 per GB in and $0.01 per GB out, when the client is not in the mount target's Availability Zone. Same-AZ access does not add that transfer. It is not the Infrequent Access retrieval charge, and it stacks with it when the reader is in the wrong zone.

Small files are a storage distortion, not a retrieval one. Infrequent Access and Archive bill a minimum file size of 128 KiB. A tree of tiny files is charged as if each were 128 KiB. That overhead can erase the storage win. It does not, by itself, create the per-GB read charge. We put it on the scorecard so a team does not "fix" retrieval and leave the overhead, or the reverse.

Archive is the colder class. The FAQ example prices Archive reads and Archive tiering higher than Infrequent Access, at $0.03 per GB in that N. Virginia walkthrough. Archive also has a 90-day minimum storage duration. Deleting or moving bytes out early bills the remainder. A lifecycle policy that rushes data into Archive, followed by a weekly reader, pays the Archive access rate on every pass. We do not recommend Archive for data a job still opens.

What we change, and what we leave

The default "stay in Infrequent Access when read" policy is why retrieval repeats. Every full read pays again, and the bytes stay on the cheap storage rate. Turning on transition back to Standard on first access changes the shape. The read tiers the bytes back, which the billing notes treat as a tiering charge, and the internal Standard timer starts. A weekly reader then keeps the working set in Standard, because each read resets that timer. Retrieval becomes a bounce, and the storage rate goes back up. That is the right trade when the data is not actually cold. It is the wrong trade when one annual audit reads the tree and the rest of the year is idle.

We name the readers before we edit the policy. A reader that must see every byte, every day, means that tree does not belong in Infrequent Access. A reader that runs once a quarter belongs in the scorecard as an expected retrieval, next to the storage savings, so finance is not surprised. A reader nobody can name is the finding. We do not turn lifecycle off for the whole file system because one prefix is hot.

AWS cost anomaly detection is how the line gets an owner. The anomaly is a jump in IADataAccess-Bytes without a matching jump in stored bytes. That pattern is a new reader, not new data. The two-week read we do produces a scorecard: file system, throughput mode, Region rates we actually used, top readers by time window, and the policy change or the accepted retrieval. It does not produce a generic "move everything to Infrequent Access" recommendation.

Send the EFS lines and we will separate storage from retrieval before anyone edits the lifecycle policy. Contact Rutagon at rutagon.com/contact, 907-841-8407, or contact@rutagon.com.

Frequently Asked Questions

Does listing a directory incur Infrequent Access retrieval?

No. AWS states that metadata operations such as listing a directory do not count as file access, and metadata stays in Standard. Retrieval on IADataAccess-Bytes is bytes read from or written in the Infrequent Access class. A name-only walk is not that meter. Opening the files is.

Why did the bill rise after storage fell?

Lifecycle lowered the storage rate and left the files in place. A later full read pays Infrequent Access data access per GB and, on Elastic Throughput, the throughput charge on the same bytes. In the FAQ's N. Virginia example those are $0.01 and $0.03 per GB. One pass can cost more than a month of Infrequent Access storage.

Does Provisioned Throughput remove the retrieval charge?

It removes the Elastic per-GB throughput line, because you are paying for provisioned performance instead. The Infrequent Access access charge still applies to bytes read in that class. Check the throughput mode and the usage types before adding the two rates together.

Should files move back to Standard on first read?

Only if the data is read often enough that repeated retrieval exceeds Standard storage. The default leaves files in Infrequent Access after a read, so every pass pays again. Transition on first access stops the repeat and puts the bytes back on the Standard rate. We pick per tree, after we name the reader.