Re: [systemd-devel] oomd wake-up frequency
On 8/25/23 7:54 AM, Michal Koutný wrote: I think the loop's event source could be disabled when no cgroups require swap monitoring [1] (and enabled lazily when such are configured). FYI I implemented your design this afternoon in case you have some time to review. https://github.com/systemd/systemd/pull/29011/commits/4806e203aa64bca942a780c4d92ea0b6b51ddd5c Thanks! -- Christian
Re: [systemd-devel] oomd wake-up frequency
Thanks for responding! Comments inline. On 8/25/23 7:54 AM, Michal Koutný wrote: Do you have any "Swap Monitored CGroups:" in output of `oomctl dump`? "Swap Monitored CGroups:" appears to be empty on the three physical machines I have at my disposal to test. (Fedora 38, Fedora 39, and GNOME OS). I think the loop's event source could be disabled when no cgroups require swap monitoring [1] (and enabled lazily when such are configured). That sounds like a great idea! Not sure whether/how much SWAP_INTERVAL_USEC could be increased to retain responsiveness. Given the empirical evidence here of all these machines not having swap monitoring to begin with, I think the interval is less of an issue compared to dynamic event sources. As it were, things like thermald are waking up every 4 seconds when the underlying hardware doesn't support async notification. It would be fantastic if we had a poll() mechanism that would coalesce the sorts of wake-ups across processes. -- Christian
Re: [systemd-devel] oomd wake-up frequency
Hello. On Tue, Aug 22, 2023 at 01:59:52PM -0700, Christian Hergert wrote: > The primary thing I see showing up when profiling an idle system is oomd. My > casual reading through the code would lead me to believe it's waking up a > CPU every .15 seconds. That coincides with swap monitoring timer. > Is there a way we could have this wake up less? My goal here is to iron out > all the little things which are causing energy drain when idle. Do you have any "Swap Monitored CGroups:" in output of `oomctl dump`? I think the loop's event source could be disabled when no cgroups require swap monitoring [1] (and enabled lazily when such are configured). Not sure whether/how much SWAP_INTERVAL_USEC could be increased to retain responsiveness. HTT, Michal [1] https://github.com/systemd/systemd/blob/1925f829ab17cee7d65cc8c350d8281f8f41588e/src/oom/oomd-manager.c#L375 signature.asc Description: PGP signature
[systemd-devel] oomd wake-up frequency
Hi! Over the past few months I've been working on improving GNOME's whole-system profiling story in the form of Sysprof¹. The primary thing I see showing up when profiling an idle system is oomd. My casual reading through the code would lead me to believe it's waking up a CPU every .15 seconds. Is there a way we could have this wake up less? My goal here is to iron out all the little things which are causing energy drain when idle. -- Christian
