Hacker News Reader

Comment by api | Hacker News Reader

Never ascribe to conspiracy what can equally well be explained by stupidity... or in this case a broken scheduler.

Broken or optimized for a different set of processors?

Don't forget that AMD can submit changes directly to the Linux Kernel but has to request or help Microsoft to make changes to Windows.

Oh I'm sure they're optimized for Intel, but that's easily explained by Intel's near-monopoly for the past 10 years. Now that AMD is gaining market share again we'll probably see this change. MS doesn't want to look bad in benchmarks.

We saw a similar situation when AMD released the Bulldozer line of CPUs and it took a minute for Microsoft to release a Hot Fix to address the issue.

https://www.anandtech.com/show/5448/the-bulldozer-scheduling...

The reality is that Windows doesn't recognize the new CPU and so it's not using a scheduler optimized for it. Where as AMD could have upstreamed any necessary changes for Linux to fully support it months ago.

And also we should keep in mind that in Kernel level improvements, Linux by far is agiler than Windows. Windows process scheduler does not have fairness for ages now. When was CFS was added to Linux? a decade?

Linux isn't free of performance regressions with new architecture. There were patches last year to better support Epyc's NUMA.

https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-n...

It just seems that Linux avoided any regressions this year but Microsoft didn't.

I wouldn't call it a conspiracy. But why would M$ optimise for NUMA, or AMD when 90% of their use cases and customer have no use of it?

Because 10% of their used base is about the same size as a first-world country.

Reply on news.ycombinator.com