Lluís Revilla

How much does cost 💲💲 CRAN?

If I am reading this right PyPi #python cost more than 1.8M a month without accountin volunteer and paid employees. Most of it is in the cloud, which #rstats equivalent would be cloud.r-project.org (which I think Posit pays?)
There are also 3 people leading and as CRAN there are other people (but CRAN manually reviews all submissions!).
I think that having mirrors distributed across universities lowers costs but just guessing.
https://dustingram.com/articles/2021/04/14/powering-the-python-package-index-in-2021/

What does it take to power the Python Package Index?

The time, money and services it takes to power PyPI in 2021

Aug 9, 2024 at 12:35pmWeb
Show thread…might work for coffee…Aug 9, 2024

@Lluis_Revilla
Might be worth to look into debian or other projects, to see how much hosting is. As Debian and other distros have much data and distribute it to those structures.

And the number will always explode once we count volunteer hours and pay them even minimal/symbolic wage. Thanks guys, you run our infrastructure!

Show threadMedian HeadroomAug 9, 2024
@Lluis_Revilla Was slightly surprised to learn that PyPI has 560000 packages, "only" 28 times more than CRAN's 20000. There's a lot more binaries on PyPI [more than a terabyte of Tensorflow binaries alone! 22TB in total! https://pypi.org/stats/ compared to half-a-terabyte of CRAN], but CRAN performs its own builds and runs a lot of checks [PyPI packagers build their own wheels]. CRAN also must have less frequent downloads.
Statistics

The Python Package Index (PyPI) is a repository of software for the Python programming language.

PyPI
Show threadLluís RevillaAug 10, 2024

@median_headroom Ouch, 1TB dedicated to only one package. Now I get how the fees can be so high.

But giving that estimation then with 22 times smaller (If I recall that CRAN was 1TB with binaries from a comment from Dirk), then the price might be around 80K€ if we assume the same number of request to the central CRAN server.
Probably it is around half of that between less downloads and the mirrors.