Edit: Clickbait…
I don’t even need to read the article to know this headline is 100% bullshit.
Doubt.
oh no, explain this then https://www.top500.org/lists/top500/list/2023/11/
The scheduler is limited but it can still schedule across all the threads and cores in a given system. It’s just doing it less efficiently. The headline is misleading.
The article itself isn’t very clear. It took me a while to figure out the scheduler algorithms were affected rather than the kernel as a whole.
It’s good then that we have choice of schedulers :)
Supercomputers are usually just a lot of smaller computers that happen to be connected with very efficient networking. Then you use something like MPI to simulate a big pool of shared memory.
Different kernel?
Pure misinformation.
https://news.ycombinator.com/item?id=38260935
tldr it’s a clickbait title.
I have already seen this horrible headline posted (and deleted) from THIS community earlier today.
Me: So Linux has had this problem since the late 90s?
Brain: Try that again old man.
Could someone please summarize this in simple terms?
The task scheduler in the Linux kernel was using a max value of 8 in its calculations rather than the actual number of cores in the machine.
The machine could have 128 cores, but the scheduler would base its calculations on 8 cores. As a result, the processes on the machine would run for less time than they should.
Damn, I hope it’s, like the other commenter said, wrong.
Claim is wrong, disregard.
Thank you
This reminds me of code I’ve written in the past and reviewed years later: At first glance it looks like it is wrong. Especially if magic numbers are involved. Then I start to think about it (hopefully with some hints in the comments 😉) and remember soon that I spent a lot of time thinking about this specific line back then and wrote it fully intentional to limit the effect of variables in my calculations 😁
Huge if true.
