A lifecycle rule would move 400 million log objects averaging 6 KB each out of standard object storage into an archive class whose per-GB rate is about 95% lower. Roughly what does the move cost and what does it save per month at 2026 US list prices?
Show the full answer Hide the answer
The assumptions, stated
400 million objects at 6 KB is about 2.4 TB, call it 2400 GB. 2026 US list prices: S3 Standard about $0.023 per GB-month, Glacier Flexible Retrieval about $0.0036 per GB-month, lifecycle transition requests into Glacier classes about $0.05 per 1000. The archive classes also carry a documented 40 KB of per-object overhead: 32 KB of index and name data billed at the Glacier rate and 8 KB billed at the Standard rate.
The arithmetic
Today: 2400 GB at \(0.023 is about **\)55 a month**.
After the move, the bytes themselves cost 2400 GB at $0.0036, about $8.60. Then the overhead:
- 400M objects x 32 KB is about 12800 GB at the Glacier rate, about $46 a month
- 400M objects x 8 KB is about 3200 GB at the Standard rate, about $74 a month
Total after archiving: roughly $129 a month, more than double the $55 it replaced. The one-off transition charge is 400 million requests at \(0.05 per 1000, about **\)20000**. The cheapest version of this project spends twenty thousand dollars to raise the monthly bill.
The number and what it rules out
Archive tiers price per object as well as per byte, so they are priced for large objects. At 6 KB the 40 KB of overhead is nearly seven times the payload. The break-even is around 40 KB per object, and the tier only behaves as advertised somewhere above 1 MB.
The assumption that dominates the error is object count, not total size. Halving the bytes changes nothing; halving the object count halves two of the three lines.
What to change first
- Compact before tiering. Roll each day's logs into one object: 2400 GB becomes about 2400 objects of 1 GB. Overhead falls to about 96 GB, transitions cost about $0.12, and the monthly bill lands near $11. The same programme now saves 80% instead of losing money.
- Expire rather than archive anything with no retention obligation. Deletion is free and total.
- Decision rule: compute overhead bytes before any lifecycle rule. If object count x 40 KB is of the same order as total size, the fix is compaction and tiering comes second.
When this is the wrong answer
When a retention rule requires each record to be individually immutable and independently retrievable, you cannot compact, and the honest answer is that archive classes are the wrong tool: use a cheaper standard-class tier or an infrequent-access class with no per-object overhead. The arithmetic also flips for large objects, where archive tiers are exactly what they claim to be and the overhead is a rounding error.