Reimplementing A Linux Rust Scheduler In eBPF Shows Very Promising Results

[phoronix]

Phoronix: Reimplementing A Linux Rust Scheduler In eBPF Shows Very Promising Results

NVIDIA software engineer Andrea Righi has implemented his "scx_rustland" Linux Rust scheduler within eBPF for very promising performance results...

https://www.phoronix.com/news/Linux-Rust-Sched-To-eBPF

[Quackdoc]

wish we could just use rust for the ebpf parts too, but some is still better then none

[V1tol]

wish we could just use rust for the ebpf parts too, but some is still better then none

You surely can and there are even some libraries exist but I wonder that too - why even bother porting when you could recompile

[ptr1337]

Andrea is working together with us on bpfland very much, and it is really great to see how good this scheduler nowadays perfoms.
I have been using bpfland already since awhile as default scheduler on our buildserver as well as on my desktop and its in a really good state.

Also, the performance in games is quite good, specially the FPS stability.

[varikonniemi]

Seems like maybe sched_ext should not be rushed into the kernel, as it's worth as anything but prototyping tool is questionable?! Maybe Linus should listen to his subordinates and let it stay out still.

[ptr1337]

Seems like maybe sched_ext should not be rushed into the kernel, as it's worth as anything but prototyping tool is questionable?! Maybe Linus should listen to his subordinates and let it stay out still.

How do you get to this conclusion ?

[duby229]

Seems like maybe sched_ext should not be rushed into the kernel, as it's worth as anything but prototyping tool is questionable?! Maybe Linus should listen to his subordinates and let it stay out still.

If you play games on Linux then this is totally worth it. There are already at least 2 schedulers optimized for gaming loads. And it isn't a prototype, it's a real thing. Pluggable schedulers is a dream come true. One for when you're gaming, one for when you're compiling, one for when you're moving data locally, one for when moving data remotely. The possibilities are endless.

This stuff here is really cool and isnt possible in kernel

[Errinwright]

Seems like maybe sched_ext should not be rushed into the kernel, as it's worth as anything but prototyping tool is questionable?! Maybe Linus should listen to his subordinates and let it stay out still.

Heaven and earth is moved to rush marginally impactful code into the kernel, yet you'd advocate not doing so for something as significant as this?

[Quackdoc]

Seems like maybe sched_ext should not be rushed into the kernel, as it's worth as anything but prototyping tool is questionable?! Maybe Linus should listen to his subordinates and let it stay out still.

I don't know what you read, but I would try reading again

[pWe00Iri3e7Z9lHOX2Qx]

Seems like maybe sched_ext should not be rushed into the kernel, as it's worth as anything but prototyping tool is questionable?! Maybe Linus should listen to his subordinates and let it stay out still.

My reaction was more, "bloody hell, can they get sched_ext merged yet?". People are creating some impressive stuff utilizing it, and I want it for gaming.