Why it matters
Workloads such as analytics scratch space, ML training data and build caches can spend most of their S3 bill on GET and PUT requests.How nOps finds it
nOps reads the last 7 complete days of S3 billing. A bucket is flagged when it was billed for on-demand PUT/LIST and GET requests on all 7 days and holds only Standard storage. The saving is the request cost times the difference between the Express One Zone and Standard request prices, minus the extra storage cost at the Express One Zone storage price, using public prices for the region, scaled to a month. It must be at least $25 a month. Express One Zone’s per-GB charge on the part of each request above 512 KB is not counted.Before you start
Applications that can use a directory bucket and its zonal endpoint, compute in the same Availability Zone, and permission to create directory buckets (s3express:CreateBucket).
How to implement
- Check that the data is request-heavy because of many small objects (under 512 KB), and that it can live in a single Availability Zone.
- Create a directory bucket in the Availability Zone where your compute runs (S3 → Directory buckets → Create bucket).
- Copy the data, or change the workload to write to the new bucket.
- Update applications to use the directory bucket. They may need a newer SDK and session-based authentication.
- Keep a copy in a regional bucket if the data must survive an Availability Zone outage.
Risks and considerations
- Express One Zone stores data in one Availability Zone, so an outage there makes the data unavailable.
- Large objects pay per-GB upload and retrieval charges that this estimate does not include.
- Directory buckets do not support every S3 feature, such as some lifecycle and replication options.
- To roll back, point the workload back at the original bucket.