Performance-Boosting MGLRU Patches Updated Against Current Linux 5.19 State

[phoronix]

Phoronix: Performance-Boosting MGLRU Patches Updated Against Current Linux 5.19 State

While there are many exciting new features coming with Linux 5.19, one of the features that wasn't submitted this cycle unfortunately was the Multi-Gen "MGLRU" code led by Google. As covered in several prior Phoronix articles, the MGLRU support has exciting performance implications for making the Linux kernel's page reclaim code far more efficient...

https://www.phoronix.com/scan.php?page=news_item&px=MGLRU-v12-For-Linux-5.19-rc

[Danny3]

I am waiting for somebody to update the Zstd code in the kernel to the latest upstream version, but this would be nice too.
I hope it lands in 5.20

[MastaG]

I also hope it lands in 5.20

[V1tol]

I also join "i hope it lands soon" movement

[guspitts]

Is there any reason for this patch set to not make it to 5.19? Discussions of needed changes for it to be accepted into mainline?

I am a happy user of the zen kernel, and it works well. From the available benchmarks it seems to have only positive and little to no down side. What's holding up that feature from being merged?

[aleksamagicka]

Is there any reason for this patch set to not make it to 5.19? Discussions of needed changes for it to be accepted into mainline?

I am a happy user of the zen kernel, and it works well. From the available benchmarks it seems to have only positive and little to no down side. What's holding up that feature from being merged?

5.19 merge window is already over, no new features are going to land in it. Also, looks like it still had bugs, judging by the article.

[bezirg]

I think this MGLRU may make the swap mechanism (coupled or not with zram or zswap) finally usable for me. Up until now, I didn't have good past experiences with swap partition (or swapfile) to the point that I ended up disabling using swap altogether.

[Vorpal]

For me it has been a few decades since memory was a problem. I threw money at the problem, getting 16 GB RAM in 2012 (that was a lot back then) and when I upgraded my computer last year: 32 GB. I run the zen kernel, but predictably I don't see a difference in day to day usage with this.

MGRLU is a way more elegant solution though and benefits devices that can't or won't be upgraded with more RAM. I look forward to what it can do on my headless Pi 2 that I still use for example.

[ResponseWriter]

Whatever it does, I don't want to see the following messages in my kernel logs again:

[332088.433728] oom-kill:constraint=CONSTRAINT_NONE,nodemask=(null),cp uset=/,mems_allowed=0,global_oom,task_memcg=/user.slice/user-1000.slice/[email protected]/session.slice/plasma-kwin_wayland.service,task=kwin_wayland,pid=7596,ui d=1000
[332088.433811] Out of memory: Killed process 7596 (kwin_wayland) total-vm:7527744kB, anon-rss:595252kB, file-rss:0kB, shmem-rss:1994884kB, UID:1000 pgtables:6572kB oom_score_adj:200

Killing Thunderbird or libreoffice, for instance, might be an inconvenience to the user but it's far preferable to taking out the entire session and everything in it.

[guglovich]

In home tasks, has anyone made any progress? In games?
I tried running 24 windows of a Lineage 2 game for a test, it used to be a working task in that game. I didn't notice any difference when comparing Linux Vanilla (Arch) and Linux Zen.

It's worth noting though, in the case of Arch kernel (Vanilla + Zen) I noticed RAM leaks which are not present in Lqx kernel. I'll have to check where else this problem exists.

P.S. I don't use SWAP and zSWAP. Is this patch not suitable for me?