Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

psi-monitor: Raise default memory pressure kill threshold to 40% for EOS 5.1 #406

Open
wants to merge 7 commits into
base: eos5.1
Choose a base branch
from

Conversation

starnight
Copy link
Contributor

Backport raising PSI monitor's default memory pressure threshold to 40% to EOS 5.1.

https://phabricator.endlessm.com/T35318

dbnicholson and others added 7 commits April 26, 2024 15:41
The more proper way to do this would be to determine them in configure,
but for such a basic C program I think we can just blindly turn them all
on and error on warnings.

https://phabricator.endlessm.com/T35318
Very basic `getopt_long` usage to allow adding options.

https://phabricator.endlessm.com/T35318
Show the current pressure, the threshold and the recovery time when
triggering an OOM to give more of a clue about what's happening.

https://phabricator.endlessm.com/T35318
Current PSI monitor's default memory pressure threshold is 10%. However,
after testing on 3 low end systems, it seems that this threshold is far
too low. You can observe 10% full memory pressure still with a decent
amount of RAM and zram available, with only minor observable effects on
the UI.

According to my recollection, 10% was quite generous before, but
over the years this seems to have changed, perhaps due to improved
ways the kernel measures things, improved behaviour under load, and
innovations like multi-gen LRU.

Experimenting and pushing these devices to the max, it seems like 40%
is a decent threshold now, consistent over 3 devices tested. At 35% you
are seeing some mouse lag but can still close apps without real
difficulty. At 40% you start to lose the ability to use the mouse.

https://phabricator.endlessm.com/T35318
Jian-Hong pointed out that I put the wrong number in the previous
commit.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants