I helped set up the first meeting between a Microsoft executive and Thomas Preston-Werner.
One of the moments that stood out to me was when Robert Youngjohns (the exec) asked Tom what it would take to have GitHub move to Azure. I was surprised that Tom had a response ready, saying that IOPS were really important and that virtual disks weren’t fast enough.
AWS gets a lot of money from both the unclassified government work they do plus the more secretive work. If they were required to disclose as part of that, they would disclose - why put all those contracts at risk?
It’s filling in a lot of the holes, but it’s putting a very convincing paper cover over the ones it misses. So it’s very hard to find the ones it didn’t fill, better hope your most valuable customers don’t walk over the paper ones!
I am surprised it is that low. The Bun Zig-to-Rust AI port was 6755 commits in like two weeks. If you make 10 commits per working day, that is 2500/year.
While that is (hopefully) the upper end of the distribution, several companies have loudly encouraged engineers to light tokens on fire to the AI gods, so it only takes a handful of the devout to push up the average in gas town like ventures.
I suspect there is a cacophony of work that happens when a commit hits the server. That request needs to get replicated, git repositories need to be repacked, pull requests need to calculate diffs, CI jobs need to execute, on and on.
That's also just assuming the good-faith usage. There are probably plenty of adversarial and poorly behaved scrapers that are putting additional load on the system.
That commit count alone should not become any problem for infrastructure, even for Azure. They probably developed some ungodly mess with actions that did not/could not translate very well on Azure infrastructure.
Seems very reasonable, from my use. I commit much more often, as checkpoints, with branch rules that prevent force pushes/deletions, so the agents can't delete anything. And, suspect MS is only counting commits, and not the eventual squashes to one commit.
We had it internally with our teams that open a PR to then push like 10-20 more commits but never actually interested in the client builds etc. turned out they opened the PR as a checkmark/ way to share the current state.
We set cooldowns and auto cancel for the ci. And then there is the developer who uses the CI compute to run tests instead of running them locally for various reasons.
We had to remind that compute isn’t for free.
Nope. You can configure CI to not run for every commit of every branch (seems insane to have full CI for every commit, unless you don't allow your devs to push until done with something, which also seems insane).
If there’s one thing that surprised me at AWS during my time there - over a decade ago now - that I was not clearly expecting in advance, it was the scale and competence of the units fulfilling the colossal and unceasing growth in capacity demand.
This led me to reconsider Amazon as a whole, and I still think of it basically as a logistics firm, with the shop and the public cloud merely a monetisation thereof.
I'm not sure if it was entirely true, but there are stories that after Microsoft bought Hotmail in the mid-90's, they quickly attempted to move them from FreeBSD (?) to Windows NT. But it failed miserably, and they went back to the original stack for another ~decade.
- the worst infinite scroll I've ever seen making it impossible to access the footer
- the title tag doesn't seem to work properly (just shows the URL in the tab title, on Chrome and Firefox)
- 2007-style keyword stuffing in meta keywords
- the entire page is client-side react with a completely empty body?
The agency that built it even proudly states on their website that they vibecode everything: https://gradientnoise.com/
EDIT: Turns out, the articles are mostly AI-generated as well? https://blog.ryanmerket.com/how-i-built-runtimewire-a-one-pe...
One of the moments that stood out to me was when Robert Youngjohns (the exec) asked Tom what it would take to have GitHub move to Azure. I was surprised that Tom had a response ready, saying that IOPS were really important and that virtual disks weren’t fast enough.
Heathen lies!
Oh, wait… we have to use it? Oh, that’s terrible…
(Probably just tea leaves. If you wanted to be extra spicy, you’d note that Jassy just threw Fable under the bus.)
So AI means 14x the checkins? That's not 14x features completed, but still... wow.
While that is (hopefully) the upper end of the distribution, several companies have loudly encouraged engineers to light tokens on fire to the AI gods, so it only takes a handful of the devout to push up the average in gas town like ventures.
Spread over a year, roughly estimating a generous 4 kbytes of data per commit, comes out to a throughput of a little under 2 MB/s.
Of course, it isn’t spread out uniformly and there is also a lot of hashing and other things going on.
Maybe pulls and clones drive more I/O ?
That's also just assuming the good-faith usage. There are probably plenty of adversarial and poorly behaved scrapers that are putting additional load on the system.
We had it internally with our teams that open a PR to then push like 10-20 more commits but never actually interested in the client builds etc. turned out they opened the PR as a checkmark/ way to share the current state. We set cooldowns and auto cancel for the ci. And then there is the developer who uses the CI compute to run tests instead of running them locally for various reasons. We had to remind that compute isn’t for free.
This led me to reconsider Amazon as a whole, and I still think of it basically as a logistics firm, with the shop and the public cloud merely a monetisation thereof.
My belief is it is likely 1% or more. And likely coming in as an avalanche.
(Probably just tea leaves.)