Microsoft brags that their repository contains 3 million files with 4 million commits.
Compare that with Google's repository which had 1 billion files and 35 million commits -- as of 2015.
On a typical workday, they commit 16,000 changes to the codebase, and another 24,000 changes are committed by automated systems.
Each day the repository serves billions of file read requests, with approximately 800,000 queries per second during peak traffic and an average of approximately 500,000 queries per second each workday.
I imagine that the windows OS repo includes IE (and edge), paint, Exploer.exe, Windows Management Systems, Cortana, and a ton of other projects that would merit their own repository in an open source community.
While separate repos do make sense for individual projects normally, as we well know, most Windows components are incredibly tightly bound together (how many different things can crash Explorer?) so a mono-repo seems to make sense for it.
I knew that Google uses version control very differently than I do, but that seems insane. That averages out to 28 new files per commit. Unless I'm bulk renaming something, I rarely have more than a few files modified, let alone added, in a commit
> I found that adorable. You [Linux] have 45,000 files. Yeah, call me when your repo starts to get big. The Windows repo has over three million files.
That Linux repo bashing at the end was very natsy.
Yes the Windows git repo is larger than the Linux git repo. Windows also has a larger scope than the Linux kernel. If you want to dick wag please bring some performance benchmarks, not footprint benchmarks.
It seems weird to use file count as a measure of size -- what is the content of those files seems more important. Moreover, is it even a good thing? 3m files _could_ (doesn't necessarily) indicate it's poorly structured/intractable. All round it just seems like a weird/unnecessary comparison to make.
The last I heard, MS imported the entire Windows history into Git. It resulted in a repo of over 300GB. To deal with this (imagine the fresh clone times...) they built a virtual filesystem on top of Git.[0]
We all know Windows is quite a mess, and many of those files will be for esoteric legacy compatibility reasons. At the same time, they've certainly done a good job of stress-testing Git!
Agreed. If they want to get into a p*ssing contest, they ought to consider they're comparing their entire OS to just the kernel. Let's pick any one distro and I bet they'll be eating their words. Then consider everything that counts as 'Linux' and I bet the daily commit count would result in Windows' development being classed as a 'critical network outage'.
I read it as a pretty fair comparison within the context. They're not comparing windows vs linux. They're comparing the challenges their repo faces vs the challenges the linux kernel repo faces. And within that context, I really hope the windows repo is substantially larger, else the linux kernel is going very very wrong.
It was a fair comparison until they started using words like 'cute', mocking the average commits per day and challenging the kernel team on size - that's when it got unnecessarily nasty. Almost condescending, Microsoft telling the guys who wrote git for their own use that they don't know how to run it.
Yeah your OS is 60 times bigger and it still has to get dragged down by business models. Nix was literally invented to be small thats why we have kernel panics. Still I'd love to get a carbon copy of the windows source. Could you imagine the power...
I don't think the author is being nasty. He's comparing the Linux kernel to Windows on the axes that are relevant here: number of files in a single repo, and number of commits per day.
The author has a clarification, but even before reading that, I didn't think the author was mocking Linux at all, just the person who tried to use the Linux repo has an example of a high-volume project.
Linux itself is both better and more important than any repo on which I'm likely to ever work, but their centralized points of control mean that I've worked on multiple repos with a higher commit-rate. It's not a great example of stress-testing an SCM product.
I'm rather surprised that Microsoft chose to use git for Windows rather than roll their own SCM. It seems to clearly not be suited for the needs of Windows. For a software company that's so invested in dogfooding, git's a really odd choice.
I know SourceSafe was a disaster and didn't think anyone used it until I started with a firm in 2013 - thankfully they were on SVN, but the first commit of the repo was from 2009, which stated simply, 'Initial import from VSS'. The company was founded in 1995...
TFVC is annoyingly proprietary but seems to follow a solid client/server model. Git support is very nice.
i don't know about solid... to me, when i can't work and have to track down someone to checkin their changes to unlock a file, there are issues. maybe other places have better practices in place to prevent this stuff from happening and i was just unlucky.
They use VSTS, that is certainly dogfooding if not anything else. As for the VCS, they probably want something distributed, and Microsoft has no distributed version control system. I'm sure there are other reasons too.
I think the amount of development required to make a new SCM was more than the amount of development required to modify git. In addition most software engineers are already familiar with git so there is less training required, and it is used as a symbol of their open source embrace.
Dogfooding does not mean that everything should be built in house. It means you should use what you’re selling.
Microsoft has built multiple source control systems. The world is moving toward distributed version control. I’m happy Microsoft is headed down that path instead of trying to invent centralized version control yet again.
This really isn't about pull requests getting completed at the exact same time AFAICT, it's about 1 person attempting to accept a pull request around the same time as 1 or more people and based on how many changes you made it you may "win" the pull request race or "lose" and have to start over.
I don't know why we as an industry can fetishize shaving milliseconds off of interaction, response, or compile time, then turn around and have dick-measuring contests about how poorly we can manage version control.
Those sound like two sides of the same coin to me?
Microsoft is bragging about shaving milliseconds off source control time (some of which has been merged upstream into main git, most of which is in GitFS which they've made available, [ETA: this particular bit is in VSTS which they've not made available as an open source effort, but also may have further smaller audience than GitFS]), and the dick-measuring here is how that reads in direct comparison to why they needed to invest in shaving those milliseconds off. Six of one, half-dozen of the other.
Edit: previous comment was toxic, and therefore downvoted/flagged. Other people above me have expressed opinion similar to mine in a more civilized manner. I am not deleting this comment to remind myself that i should keep my comments civil even if i have a strong emotional reaction to something someone wrote.
I could never work on Windows. I'd just want to burn the whole thing to the ground and start over. There is no way an OS actually needs to be that massive; as helpfully illustrated by the comparison to Linux's metrics at the end.
Kernel versus vast monorepo of most of the OS, many of its pack-in apps, etc. I've heard it said the core NT Kernel itself is mostly on par with the Linux Kernel, it's when you consider all the other things that are a part of Windows when it gets that huge. (Also why much of the work in GitFS the team did was making sure that partial checkouts work well, because few people need a work tree of the entire thing at once.)
I'm really confused by this monorepo thing. So many people have gleefully hopped on to this particular cult, and the most reasonable reasoning that I've heard is it helps to glob together interdependent parts of a system, but I feel like that's throwing away everything we take for granted about what makes good software: that it's better to separate your code into distinct pieces with well-defined scope (and their own version numbers that reflect their interfaces); to expose your interfaces, rather than your implementations, and if someone is dependent on the exact implementation of module B then maybe your design is fucked and you should start over before you hurt someone.
So I'm skeptical of that first argument. What does this do for you that a decent architect doesn't do better? Is it a way for dangerously lazy project managers to justify themselves? Is it because it works for Google so of course it works for everyone? Is it a workaround for Github's private repo limit but nobody wants to admit it because they're afraid Github will figure it out? If Windows needs this, how the hell are Debian, Fedora and Ubuntu doing all these releases? Do you know how many repositories they have to deal with?!
Amen. Bragging about having a large repo is showing the world that you can't design large software systems.
Just like creating branches in Git is cheap, creating repos in Git is just as cheap. The difference is that you need actual design and engineering discipline to version your interfaces and manage those dependencies.
Having a kitchen-sink repo is poor engineering, and massive code smell.
37 comments
[ 4.6 ms ] story [ 87.2 ms ] threadhttps://www.visualstudio.com/learn/race-to-push/
Compare that with Google's repository which had 1 billion files and 35 million commits -- as of 2015.
On a typical workday, they commit 16,000 changes to the codebase, and another 24,000 changes are committed by automated systems.
Each day the repository serves billions of file read requests, with approximately 800,000 queries per second during peak traffic and an average of approximately 500,000 queries per second each workday.
https://cacm.acm.org/magazines/2016/7/204032-why-google-stor...
Microsoft's repo in question here is only for the WINDOWS OS.
Google has all their code in one repository. And it's not git.
Microsoft's repo may well be the biggest git repo.
That Linux repo bashing at the end was very natsy.
Yes the Windows git repo is larger than the Linux git repo. Windows also has a larger scope than the Linux kernel. If you want to dick wag please bring some performance benchmarks, not footprint benchmarks.
Edit: missed "indicate".
We all know Windows is quite a mess, and many of those files will be for esoteric legacy compatibility reasons. At the same time, they've certainly done a good job of stress-testing Git!
[0]https://techcrunch.com/2017/05/24/microsoft-now-uses-git-and...
Linux itself is both better and more important than any repo on which I'm likely to ever work, but their centralized points of control mean that I've worked on multiple repos with a higher commit-rate. It's not a great example of stress-testing an SCM product.
TFVC is annoyingly proprietary but seems to follow a solid client/server model. Git support is very nice.
Microsoft has built multiple source control systems. The world is moving toward distributed version control. I’m happy Microsoft is headed down that path instead of trying to invent centralized version control yet again.
Microsoft is bragging about shaving milliseconds off source control time (some of which has been merged upstream into main git, most of which is in GitFS which they've made available, [ETA: this particular bit is in VSTS which they've not made available as an open source effort, but also may have further smaller audience than GitFS]), and the dick-measuring here is how that reads in direct comparison to why they needed to invest in shaving those milliseconds off. Six of one, half-dozen of the other.
Edit: previous comment was toxic, and therefore downvoted/flagged. Other people above me have expressed opinion similar to mine in a more civilized manner. I am not deleting this comment to remind myself that i should keep my comments civil even if i have a strong emotional reaction to something someone wrote.
So I'm skeptical of that first argument. What does this do for you that a decent architect doesn't do better? Is it a way for dangerously lazy project managers to justify themselves? Is it because it works for Google so of course it works for everyone? Is it a workaround for Github's private repo limit but nobody wants to admit it because they're afraid Github will figure it out? If Windows needs this, how the hell are Debian, Fedora and Ubuntu doing all these releases? Do you know how many repositories they have to deal with?!
Just like creating branches in Git is cheap, creating repos in Git is just as cheap. The difference is that you need actual design and engineering discipline to version your interfaces and manage those dependencies.
Having a kitchen-sink repo is poor engineering, and massive code smell.