Reading large objects from Glacier Instant Retrieval

[IdrisMiles]

Hi there,

I'm curious about the potential cost implications of reading large objects in the Glacier Instant Retrieval tier with mountpoint-s3.
If I'm understanding the this part of the docs correctly about configuring performance for parallel requests, if you were to try and read for example a 1GB object residing in Glacier Instance Retrieval with the default 8 MiB --read-part-size value. Would that essentially incur x128 1GB data retrievals? where 1024 / 8 = 128.
Or is S3 only charging for the data retrieval once?

[dannycjones]

Hey @IdrisMiles, thanks for your interest!

To address the specific example, you would be charged once for the 1GB of data retrieved (alongside per-request GET cost).

If you were to use a smaller read part size, the number of requests would increase however the amount of data read would remain the same. For pricing, the per-request cost would increase but the data retrieval cost would remain the same. A larger object would incur an increase in both pricing elements, as the amount of data retrieved would exceed 1GB and the number of 8MB requests to retrieve the data would also increase.

If you wish to reduce costs here, you could explore increasing the read part size for the Mountpoint file system, possibly at the expense of worsened read performance.

Do let me know if there's still any ambiguity and I can reopen!

[IdrisMiles]

Hi @dannycjones thanks for clarifying, this answers my question. Data retreival is charged once regardless of the number of GET requests from different read part sizes.
Understood that the quantity of GET requests also has a charged associated as per the S3 pricing.