Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]


Groups > linux.debian.kernel > #68560 > unrolled thread

Bad desktop performance with Linux 5.8

Started byKévin Le Gouguec <kevin.legouguec@gmail.com>
First post2020-11-14 23:30 +0100
Last post2020-12-06 16:10 +0100
Articles 4 — 2 participants

Back to article view | Back to linux.debian.kernel


Contents

  Bad desktop performance with Linux 5.8 Kévin Le Gouguec <kevin.legouguec@gmail.com> - 2020-11-14 23:30 +0100
    Re: Bad desktop performance with Linux 5.8 Georgi Naplatanov <gosho@oles.biz> - 2020-11-15 00:50 +0100
      Re: Bad desktop performance with Linux 5.8 Kévin Le Gouguec <kevin.legouguec@gmail.com> - 2020-11-15 15:30 +0100
      Re: Bad desktop performance with Linux 5.8 Kévin Le Gouguec <kevin.legouguec@gmail.com> - 2020-12-06 16:10 +0100

#68560 — Bad desktop performance with Linux 5.8

FromKévin Le Gouguec <kevin.legouguec@gmail.com>
Date2020-11-14 23:30 +0100
SubjectBad desktop performance with Linux 5.8
Message-ID<Bbcdr-76a-1@gated-at.bofh.it>

[Multipart message — attachments visible in raw view] — view raw

Hello,

I am running Debian Buster (XFCE) on a Dell Laptop[1] with some packages
upgraded from backports[2], notably linux-image-amd64.  After upgrading
to version 5.8, I started noticing some UI "choppiness".  To give a few
examples:

- when typing (in any application, e.g. Emacs, Terminator, Firefox): at
  its worst, I can type a bunch of words, sit back, wait a few seconds,
  then watch the words appear letter by letter.

- when launching programs: startup times are very inconsistent; e.g. if
  I loop on
    while time emacs -Q --eval '(kill-emacs)' ; do continue ; done
  or
    while time terminator -x "bash -c exit" ; do continue ; done
  times vary between 1× and 3× what I observe on kernel 5.7 (where times
  barely vary at all, ±5% at most).

- when scrolling a PDF with Evince with the touchpad: the display
  "freezes" (the pages do not seem to be scrolling) and jumps to the new
  position after a few seconds; nonetheless, the pages *do* scroll
  continuously, as the mouse pointer changes from an arrow to an I-beam,
  presumably as text and whitespace move past it.

Neither htop, journalctl nor dmesg report anything suspicious to my
untrained eye; the only difference I spotted was the CPU governor: on
5.7 it is "powersave"; on 5.8 it is "ondemand".  Naively cpufreq-setting
all cores to powersave on 5.8 does not make the problem go away, though.

One curious thing is that looking at
    watch -n0.5 grep MHz /proc/cpuinfo

- on 5.7, the system is very responsive, and the idle frequency is
  around 800 MHz;
- on 5.8, the system lags despite the idle frequency never going below
  1700 MHz.

This might all be a red herring though, for all I know.  I did not
immediately realize that the performance drops came from 5.8; I only
understood that after studying a bunch of other hypotheses (uninstalling
xserver-xorg-video-intel, disabling services such as speech-dispatcher
and tor), coming up short, rebooting, opening Grub's "advanced options"
and realizing that 5.7 was still on my system.

I've since rebooted multiple times on 5.7 and 5.8, and I'm positive that
the problem only occurs with 5.8.


I'm not sure what to look at next.  I've searched the Web for people
having performance issues with 5.8, but nothing caught my eye.

I realize that "performance bad ☹" testimonies like this are probably
not very actionable; let me know if there's anything I can do to make
this "diagnosis" more precise.  Also, I hope this is the right place for
this kind of support request; I assume it is too vague as-is for a
proper bug report…  Let me know if I should post this elsewhere.


Thank you for your time, and sorry for the poor report.


[1] Dell Latitude 3190.  Attached outputs from lscpu and lspci:

[toc] | [next] | [standalone]


#68561

FromGeorgi Naplatanov <gosho@oles.biz>
Date2020-11-15 00:50 +0100
Message-ID<BbdsS-7LB-1@gated-at.bofh.it>
In reply to#68560
On 11/15/20 12:05 AM, Kévin Le Gouguec wrote:
> Hello,
> 
> I am running Debian Buster (XFCE) on a Dell Laptop[1] with some packages
> upgraded from backports[2], notably linux-image-amd64.  After upgrading
> to version 5.8, I started noticing some UI "choppiness".  To give a few
> examples:
> 
> - when typing (in any application, e.g. Emacs, Terminator, Firefox): at
>   its worst, I can type a bunch of words, sit back, wait a few seconds,
>   then watch the words appear letter by letter.
> 
> - when launching programs: startup times are very inconsistent; e.g. if
>   I loop on
>     while time emacs -Q --eval '(kill-emacs)' ; do continue ; done
>   or
>     while time terminator -x "bash -c exit" ; do continue ; done
>   times vary between 1× and 3× what I observe on kernel 5.7 (where times
>   barely vary at all, ±5% at most).
> 
> - when scrolling a PDF with Evince with the touchpad: the display
>   "freezes" (the pages do not seem to be scrolling) and jumps to the new
>   position after a few seconds; nonetheless, the pages *do* scroll
>   continuously, as the mouse pointer changes from an arrow to an I-beam,
>   presumably as text and whitespace move past it.
> 
> Neither htop, journalctl nor dmesg report anything suspicious to my
> untrained eye; the only difference I spotted was the CPU governor: on
> 5.7 it is "powersave"; on 5.8 it is "ondemand".  Naively cpufreq-setting
> all cores to powersave on 5.8 does not make the problem go away, though.
> 
> One curious thing is that looking at
>     watch -n0.5 grep MHz /proc/cpuinfo
> 
> - on 5.7, the system is very responsive, and the idle frequency is
>   around 800 MHz;
> - on 5.8, the system lags despite the idle frequency never going below
>   1700 MHz.
> 
> This might all be a red herring though, for all I know.  I did not
> immediately realize that the performance drops came from 5.8; I only
> understood that after studying a bunch of other hypotheses (uninstalling
> xserver-xorg-video-intel, disabling services such as speech-dispatcher
> and tor), coming up short, rebooting, opening Grub's "advanced options"
> and realizing that 5.7 was still on my system.
> 
> I've since rebooted multiple times on 5.7 and 5.8, and I'm positive that
> the problem only occurs with 5.8.
> 
> 
> I'm not sure what to look at next.  I've searched the Web for people
> having performance issues with 5.8, but nothing caught my eye.
> 
> I realize that "performance bad ☹" testimonies like this are probably
> not very actionable; let me know if there's anything I can do to make
> this "diagnosis" more precise.  Also, I hope this is the right place for
> this kind of support request; I assume it is too vague as-is for a
> proper bug report…  Let me know if I should post this elsewhere.
> 
> 
> Thank you for your time, and sorry for the poor report.
> 
> 
> [1] Dell Latitude 3190.  Attached outputs from lscpu and lspci:
> 
> 
> 
> [2] Potentially relevant backport packages:
> 
>         $ aptitude search '?narrow(?installed, ?archive(buster-backports)) ?name(firmware~|linux~|wireless)' -F '%p'
>         firmware-amd-graphics
>         firmware-iwlwifi
>         firmware-linux
>         firmware-linux-nonfree
>         firmware-misc-nonfree
>         linux-image-5.8.0-0.bpo.2-amd64
>         linux-image-amd64
>         wireless-regdb
> 
>     Potentially relevant non-free packages:
> 
>         $ aptitude search '?installed ?section(non-free/) !?name(-doc$~|^manpages-)' -F '%p'
>         amd64-microcode
>         firmware-amd-graphics
>         firmware-iwlwifi
>         firmware-linux
>         firmware-linux-nonfree
>         firmware-misc-nonfree
>         intel-microcode
> 


Hi Kévin,

Linux kernel version (5.8) is old and unsupported (EOL) branch. Maybe
you should wait for 5.9 to be available from backports or even 5.10 -
this version will be LTS and next Debian stable version will probably
use it. While waiting you can try to use 5.7 if possible.

Kind regards
Georgi

[toc] | [prev] | [next] | [standalone]


#68569

FromKévin Le Gouguec <kevin.legouguec@gmail.com>
Date2020-11-15 15:30 +0100
Message-ID<Bbrcu-7ET-11@gated-at.bofh.it>
In reply to#68561
Georgi Naplatanov <gosho@oles.biz> writes:

> Hi Kévin,
>
> Linux kernel version (5.8) is old and unsupported (EOL) branch. Maybe
> you should wait for 5.9 to be available from backports or even 5.10 -
> this version will be LTS and next Debian stable version will probably
> use it. While waiting you can try to use 5.7 if possible.

Hello Georgi,

Thanks for the advice.  I figured it made sense to report this asap
because it'd be easier to find the cause of the issue if there's only
one version's worth of changelog to examine, but I can wait for future
versions.

[toc] | [prev] | [next] | [standalone]


#68779

FromKévin Le Gouguec <kevin.legouguec@gmail.com>
Date2020-12-06 16:10 +0100
Message-ID<Bj3PI-3Hy-17@gated-at.bofh.it>
In reply to#68561
Georgi Naplatanov <gosho@oles.biz> writes:

> Linux kernel version (5.8) is old and unsupported (EOL) branch. Maybe
> you should wait for 5.9 to be available from backports or even 5.10 -
> this version will be LTS and next Debian stable version will probably
> use it. While waiting you can try to use 5.7 if possible.

Quick followup since 5.9 has just hit buster-backports: the symptoms
persist.

I don't mind hanging on to 5.7 and waiting until 5.10 to debug this
further; though if anyone has ideas of things I could try to
enable/disable/reconfigure to isolate the problem, I remain all ears.

[toc] | [prev] | [standalone]


Back to top | Article view | linux.debian.kernel


csiph-web