Experimental Schedutil Patches Yield 30% Boost To Web Browser Benchmark On Linux

[phoronix]

Phoronix: Experimental Schedutil Patches Yield 30% Boost To Web Browser Benchmark On Linux

Google engineer Qais Yousef has posted a set of 16 patches for the "Schedutil" scheduler utilization code within the Linux kernel to better manage system response time. Schedutil is often used by default on many Linux distributions and with these patches a popular web browser benchmark can be as much as 30% faster with these kernel patches...

https://www.phoronix.com/news/Schedutil-30p-Speedometer-Boost

[Quackdoc]

I still don't understand why anyone ships schedutil by default. Maybe this will change it, but the only thing i've tested that schedutil doesn't flounder on android

[markg85]

I still don't understand why anyone ships schedutil by default. Maybe this will change it, but the only thing i've tested that schedutil doesn't flounder on android

Any particular reason why you say that?
I'm on arch and here the default apparently is schedutil (cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor shows schedutil) and i'm not having any issues whatsoever with it. Keep in mind that i care for a quiet pc! So a fans blazing max frequency is something i rather not have on my cpu unless i start a task that justifies it, like compilation or transcoding. But there too i find schedutil to perform just fine.

[Volta]

Any particular reason why you say that?
I'm on arch and here the default apparently is schedutil (cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor shows schedutil) and i'm not having any issues whatsoever with it. Keep in mind that i care for a quiet pc! So a fans blazing max frequency is something i rather not have on my cpu unless i start a task that justifies it, like compilation or transcoding. But there too i find schedutil to perform just fine.

Schedutils was always broken and the slowest piece of crap. Maybe now it will change, but who knows?

[dmalcolm]

Does anyone know what tool is being used to draw those "ASCII" art plots in the cover letter?

e.g. this kind of thing:

[PRE]

rampup-5088 util_avg running
┌────────────────────────────────────────────────────────────────────────┐
1015.0┤ ▄▄▄▄▄▄▄▄▄▟▀▀▀▀▀▀▀▀▀▀▀▀│
│ ▗▄▄▄▛▀▀▀▀▘ │
│ ▗▄▟▀▀▀ │
│ ▄▟▀▀ │
761.2┤ ▄▟▀▘ │
│ ▗▛▘ │
│ ▗▟▀ │
507.5┤ ▗▟▀ │
│ ▗▛ │
│ ▄▛ │
│ ▟▘ │
253.8┤ ▐▘ │
│ ▟▀ │
│ ▗▘ │
│ ▗▛ │
0.0┤ ▗ ▛ │
└┬───────┬───────┬───────┬───────┬──────┬───────┬───────┬───────┬───────┬┘
1.700 1.733 1.767 1.800 1.833 1.867 1.900 1.933 1.967 2.000
​

[/PRE]

...though that's getting horribly mangled by my attempt to quote in the forum (sorry!)

[Quackdoc]

Any particular reason why you say that?
I'm on arch and here the default apparently is schedutil (cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor shows schedutil) and i'm not having any issues whatsoever with it. Keep in mind that i care for a quiet pc! So a fans blazing max frequency is something i rather not have on my cpu unless i start a task that justifies it, like compilation or transcoding. But there too i find schedutil to perform just fine.

I have found ondemand to almost always perform better and more reliably

[SearingHeat]

Does anyone know what tool is being used to draw those "ASCII" art plots in the cover letter?

Probably gnuplot with terminal output.

[dmalcolm]

Probably gnuplot with terminal output.

Aha - thanks! Looks like it might be the "block" terminal type from http://gnuplot.info/docs/Terminals.html

[Mitch]

I thought the main problem with schedutil was that it never won in performance, efficiency, nor power usage minimization. At least this is how the story was for it vs Ondemand gov, Performance gov, and PState. I believe schedutil was designed to take in more heuristics vs other governors.

I think some devices, like Android, didn't necessarily have a PState they could use and schedutil seemed better for their hardware.

It would be really cool if we see improvements anyways, as those are bound to trickle down somewhere.

[muncrief]

I used schedutil until recently, but it disappeared at the same time gnome-disks lost the ability to read SMART information.

Now "cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_governors" just shows performance and powersave. I also tried rebuilding paramano as it's the only system tray utility I've ever found that easily allowed the governor to be changed and the current CPU core frequencies to be displayed, but it no longer builds. Of course it's very old so I figured it had to break someday.

So it looks like recent kernels (I've tried CachyOS and Arch kernels) have borked a few things and I sure am looking forward to the day things get back to normal. Not being able to easily view SMART data, and view and set CPU frequencies, is quite a set back. And yes, I know there are a plethora of command line tools to do this stuff, but they're quite clumsy and the output is difficult to parse. And the few other GUI options I've found are equally clumsy, or in the case of SMART data don't show all disks, so you have to use the command line anyway.