Talk:Improving performance

Section 5.2: liveroot hasn't been updated since 2016

The `liveroot` script in the AUR hasn't been updated since 2016, and I couldn't get it to work properly. Should we still be recommending it in this article?

—This unsigned comment is by Saltedcoffii (talk) 20:34, 19 February 2023 (UTC). Please sign your posts with ~~~~!

Someone has gotten it working as early as last year. This does looks unmaintained, but the overall approach should still work. -- CodingKoopa (talk) 17:33, 23 February 2023 (UTC)

4.5: Kernel module parameter might be needed for ReBAR on NVIDIA

Along with the standard process of updating the firmware and enabling the setting in the BIOS, you might need the NVreg_EnableResizableBar=1 module parameter to enable ReBAR on NVIDIA. https://github.com/NVIDIA/open-gpu-kernel-modules/pull/3

At least on my system, /proc/driver/nvidia/params shows this feature being disabled by default, even when enabled in the BIOS. JL2210 (talk) 03:05, 26 July 2024 (UTC)

Watchdog recommendation change

Apart from the fact that Special:Diff/826433 looks a lot like AI generated content (See this cleanup of <code> tags, which remain in Improving performance#Checking Hardware Watchdog Support) shouldn't it be better documented why the wiki's suggestion to disable watchdogs for performance suddenly becomes one to enable them? Cvlc (talk) 11:29, 25 February 2025 (UTC)

You're right—the content may be AI-generated, but that doesn’t diminish its value. The reasoning behind it appears solid and well-thought-out, and it didn’t abruptly replace the previous suggestion, which had been in place for years. That said, how was disabling the watchdog justified? There’s no clear benefit to turning it off; in fact, it seems to only eliminate the useful functionality that the watchdog itself provides. Lucenera (talk) 11:38, 25 February 2025 (UTC)

Well sorry to disagree, but I think it can definitely diminish its value. Reasoning always appears solid in AI generated content, that does not make it correct. Does ShutDownWatchdog= even exist ?

Anyway my point is that it should probably be better documented when an article abruptly changes recommendations, not that watchdogs are or aren't justified.

[edit]ShutDownWatchdog= is an obsolete alias. So yeah, sorry but AI generated edits which aren't even properly formatted and dump errors into the wiki should just be reverted. Cvlc (talk) 14:43, 25 February 2025 (UTC)

Thanks for bringing it up. Please add a new Template:Accuracy if you still see anything that has not been corrected yet.

@Lucenera: It's not proper to remove an accuracy template with (un-referenced) third-party generated instructions that "appear solid". It's not a trivial setting, if you want it. I hope you can confirm the section is accurate by testing it. --Indigo (talk) 20:47, 25 February 2025 (UTC)

The section has been fully tested on my setup: a Lenovo ThinkPad T480 with 64 GB of RAM, desktop use, two NVMe disks, 512GB and 1TB, with OpenZFS and one zpool per disk, Arch Linux with Gnome and watchdog fully active from day one, with the suggested settings. In fact, ShutDownWatchdog has been replaced by RebootWatchdogSec. Thanks for the correction. Lucenera (talk) 21:20, 25 February 2025 (UTC)

The content before my first change yesterday was definitely wrong. It read like the systemd watchdog would trigger a restart which is the opposite of how a hardware watchdog works.

When skimming the following paragraph, I also had the feeling it was AI generated and biased towards employing a hardware watchdog. It might be ok to generate AI content for usage in a wiki - but not to publish it without double-checking. I have not in-depth checked that paragraph in question. JonasKonrad (talk) 09:18, 26 February 2025 (UTC)

Thanks for the input both. I've added [1] as my 2c, but would not have thought about it without reading this talk item. You're welcome to adjust it, or flesh it out. The original case may be an example of how a generator takes settings from a 10+ year old PID1 blog as status quo, instead of parsing rolling release settings against it. Whatever the case, I just like to stress again that particularly Template:Accuracy and Template:Out of date are meant to alert readers and contributors alike.

Let's close this and roll on with improving our content. --Indigo (talk) 19:48, 26 February 2025 (UTC)

So how did you test the performance improvement if your watchdog is fully active from day one? — Lahwaacz (talk) 13:17, 28 February 2025 (UTC)

Because, before the current installation, on the same system, I had the watchdog off. Now it's not that I've improved by activating a watchdog, but neither have I worsened compared to disabling it. In short, it is absolutely irrelevant. But since it is there for a reason, it can be safely used without affecting performance in any way. The suggestion to deactivate it should simply disappear and there wouldn't even need to be a paragraph about the service, what it is for and how to use it. Lucenera (talk) 21:30, 28 February 2025 (UTC)

I've updated Power management#Disabling NMI watchdog. However I'm unsure how many interrupts should be considered enough to hurt performance. Cvlc (talk) 23:06, 28 February 2025 (UTC)

It may escape this article on performance. If I understand kernel doc right, one should only use it in conjunction with parameters like panic=-1. Perhaps another crosslink to Realtime kernel patchset/ Realtime kernel patchset#Scheduling latency is more useful. --Indigo (talk) 21:55, 1 March 2025 (UTC)