Skip to main content
Topic: Kernel 6.3 Changes Probably Should Be Reverted (Read 1207 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Kernel 6.3 Changes Probably Should Be Reverted

There used to be a very detailed, big comparison table of XZ vs ZSTD with all level switches (zstd has too many of them for the sake of having so many levels...). As far as I remember ZSTD was a huge resource hog. Very huge resource hog comparing with XZ (level number don't match so you need to use matching levels and adjust table by moving up and down in the comparison table).

Since ZSTD is a resource hog compering to XZ. Now think about comparing it to LZ4. Even worse!
Since everyone who has, lets say, 4 GiB of RAM, has only 4GiB of RAM and older hardware (CPU). So using a lot of RAM to compress RAM and a lot of CPU  (full CPU cache that is not enough - hence usage of RAM) makes absolutely no sense.

Since kernel 6.3 its using ZSMALLOC and ZSTD for ZSWAP  by itself is a resource hog and makes the kernel and system not snappy. Of course one can use GRUB* to set parameters for ZSWAP and MODULES for early load in mkinitcpio etc., but it is yet another step for configuration.

So new hardware has a lots of RAM and CPU cache, so its not a problem for it with RAM.
Old hardware has less RAM and slower CPU and less CPU cache so it has problem with RAM.
And yet someone with their  "infinite wisdom" made default for new hardware. It makes no sense.


Please revert to lz4 and z3fold by default for Artix.

Tested with two browsers and multiple tabs opened and mpv playing online video via python script (shhh....).

*(soon to be deprecated because of GPLv3 incompatibiliy -see Debian and systemdboot...)

Re: Kernel 6.3 Changes Probably Should Be Reverted

Reply #1
According to the Linux Foundation PDF papers z3fold is the best regardless of anything. So lz4+z3fold is the most optimal and shoudl be the default.

Here is the screenshot with new ZSWAP parameters on my machine:





'https://pasteboard.co/WRVn1dHuaQMg.png

P.S. Normally in Arch (systemd) I used to write udev rules for the I/O scheduler. For SSD the optimal was mq-deadline (cold application start) and for spinning BFQ (e.g. for external drive with media, formatted optimally for big files with XFS).
I can't see antything equivalant in Artix Wiki for OpenRC and Runit.

Re: Kernel 6.3 Changes Probably Should Be Reverted

Reply #2
Would you please provide some proof links for "ZSTD is a resource hog compering to XZ"? I didn't notice such thing on my 12 and 14 years old computers. Maybe I don't know where to look? Also, a link for "z3fold is the best regardless of anything" would be nice. Thanks.

Re: Kernel 6.3 Changes Probably Should Be Reverted

Reply #3
As far as I remember ZSTD was a huge resource hog.

RTFM:

https://en.wikipedia.org/wiki/Zstd
https://archlinux.org/news/now-using-zstandard-instead-of-xz-for-package-compression/
"Wer alles kann, macht nichts richtig"

Artix USE="runit openrc slim openbox lxde gtk2 qt4 qt5 qt6 conky
-gtk3 -gtk4 -adwaita{cursors,themes,icons} -gnome3 -kde -plasma -wayland "

Re: Kernel 6.3 Changes Probably Should Be Reverted

Reply #4

Nothing you shared looks like a manual even in the dark.

If you are unhappy with the defaults change the values and stop barking like a dumb dog

And yet someone with their  "infinite wisdom" made default for new hardware. It makes no sense.

Who would have thought a bleeding edge rolling release distribution would prefer new stuff instead of maintaining compatibility with legacy hardware?

 

Re: Kernel 6.3 Changes Probably Should Be Reverted

Reply #5

Nothing you shared looks like a manual even in the dark.

If you are unhappy with the defaults change the values and stop barking like a dumb dog


"It is better to light a small candle than to curse the darkness."
"Wer alles kann, macht nichts richtig"

Artix USE="runit openrc slim openbox lxde gtk2 qt4 qt5 qt6 conky
-gtk3 -gtk4 -adwaita{cursors,themes,icons} -gnome3 -kde -plasma -wayland "