Linux 5.16's New Cluster Scheduling Is Causing Regression, Further Hurting Alder Lake

[phoronix]

Phoronix: Linux 5.16's New Cluster Scheduling Is Causing Regression, Further Hurting Alder Lake

Linux 5.16-rc1 is coming out later today and already I'm seeing some fallout in the new kernel's performance... In particular, bad news for Alder Lake that is already seeing the Linux performance trailing Windows 11 seemingly due to the lack of Thread Director integration right now in the kernel and any other missing optimizations around Intel's hybrid architecture. A new feature of Linux 5.16 is unfortunately having unintended regressions for Alder Lake with at least the flagship Core i9 12900K. Here are the results from the latest kernel bisecting that uncovered this latest upstream slowdown.

https://www.phoronix.com/vr.php?view=30686

[perpetually high]

Big yikes. Great work again, Michael. I have no dog in this fight; I hope Intel gets this cleared up soon.

For those wondering:

As of last testing, this cluster scheduling doesn't have any impact for AMD processors with not yet having any topology support there.

This is the problem of growing pains with new tech, and exactly why I avoided Alder Lake. Has nothing to do about Intel or raw performance or even performance-per-watt, I did not want to deal with these issues that take away from me enjoying my new tech.

[ms178]

For those wondering, there was also a performance regression on AMD's EPYC spotted with this patchset, but AMD engineers were quick to report it and it there was some work done to prevent this from happening on AMD hardware.

[avem]

Big yikes. Great work again, Michael. I have no dog in this fight; I hope Intel gets this cleared up soon.

This is the problem of growing pains with new tech, and exactly why I avoided Alder Lake. Has nothing to do about Intel or raw performance or even performance-per-watt, I did not want to deal with these issues that take away from me enjoying my new tech.

You still can use taskset and avoid any growing pains. It's currently the recommended solution for Windows 10 (processlasso and the likes) as well but it's not been so bad in terms of utilizing the full performance of ADL in the first place. It's the first time I see Linux demonstrating such a poor performance.

Lastly, Linux must not suck as well, as we've had big-little ARM cores for ages now and Linux supports them perfectly. Probably there are some systemctl or/and boot variables which can let the kernel know which cores are which to work properly - strangely no one has researched this.

[Hans Bull]

Introduced by a Intel engineer?

[SteamPunker]

It's incredible how so few people have been supporting Michael, even though he has been single-handedly making crucial contributions to the development of Linux in the form of meticulous benchmarking and detailed reporting. This is far from the first serious performance regression his efforts have yielded! Can we finally get some more funding for this guy? Or perhaps some big sponsors like Google or Intel?

🎵 Toss a coin to Phoronix, oh Linux Community, oh Linux Community! 💰

[avem]

Introduced by a Intel engineer?

Yeah, this has caught my attention as well. Not just one, two of them.

[F.Ultra]

You still can use taskset and avoid any growing pains. It's currently the recommended solution for Windows 10 (processlasso and the likes) as well but it's not been so bad in terms of utilizing the full performance of ADL in the first place. It's the first time I see Linux demonstrating such a poor performance.

Lastly, Linux must not suck as well, as we've had big-little ARM cores for ages now and Linux supports them perfectly. Probably there are some systemctl or/and boot variables which can let the kernel know which cores are which to work properly - strangely no one has researched this.

AFAIK all big.little schedulers are proprietary and the stock Linux scheduler works just as bad on those as it does on Alder lake since there AFAIK is no one that uses that arc outside of smart phones. At the moment Intel haven't released any info on how to talk to Thread Director so it's not possible to build support in Linux until they do.

[F.Ultra]

Yeah, this has caught my attention as well. Not just one, two of them.

Yeah but they worked on Jacobsville so most likely that are several closed doors away from the Alder Lake people, Intel is huge enough that the left hand does not know that the right hand is doing.

And on Alder Lake each E-core module (4 cores per module so 2 modules per chip in the current cpu) shares L2 so with cluster scheduling enabled you have a much higher chance that you get stuck on the E-cores which is most likely the problem that Michael is seeing here.

[lethalwp]

i wonder, since it s a desktop, should nt we just ignore E cores and go full perfs?
i think these are a good idea on laptops, but a gimmik on desktops to artificially increase corecount, when intel cannot cope with amd cores?