| Wizzup | phoenix.maemo.org / leste.maemo.org is down for a few more hours today, and then they will be on UPS | 12:17 |
|---|---|---|
| inky | i want to also tell that i don't know why, but under maemo pinephone not only works longer than under postmarketos, but the handset itself is cold, under pmos it is always hot. even if on apps run. | 12:39 |
| inky | i don't know why is this, i assume you integrated all or almost all patches, so i guess this is more because of devuan vs alpine thing. maybe glibc vs musl also matter, i don't know. | 12:39 |
| inky | i hope it's not just the older versions of everything. | 12:39 |
| Wizzup | could be wayland vs X | 12:44 |
| Wizzup | or gtk4/qt6 vs gtk2, etc | 12:44 |
| uvos | inky: seams mutch more likely that you have something installed thats using cpu time on pmos rather than something we did, we dont hit any sleepstates on the pp as far as i am aware | 12:46 |
| uvos | inky: i gues running powertop and compearing would reveal who it is | 12:46 |
| Wizzup | we def. worked on pp power management | 12:47 |
| uvos | Wizzup: sure but we dont hit crust sleep at all | 12:47 |
| uvos | iirc | 12:47 |
| uvos | also on pmos you could have any interface installed or no interface installed so like Wizzup sais it might be heavy or not | 12:47 |
| Wizzup | right, the mode where you basically suspend and can't listen to music we don't hit | 12:47 |
| uvos | Wizzup: right i know but pmos can hit it depending on the installed de | 12:48 |
| arno11 | Wizzup: btw zram is still there in n900 just not launched on boot | 14:41 |
| arno11 | i tried it and as expected, it is a bit slower than swapfile | 14:42 |
| arno11 | but it works | 14:42 |
| Wizzup | yeah, so it is disabled as expected | 14:47 |
| arno11 | yep | 14:47 |
| buZz | arno11: did you try those zram tweaks perhaps? | 15:08 |
| buZz | https://wiki.archlinux.org/title/Zram#Optimizing_swap_on_zram | 15:08 |
| buZz | not sure if it'll have a lot of impact, but it might | 15:08 |
| arno11 | buZz: hi ! ah yes interesting, i'll try. thx | 15:43 |
| buZz | i'll be curious to see if such improves it for n900 | 15:44 |
| arno11 | (currently trying) | 15:55 |
| Wizzup | 'zram on n900 or not' is something that comes up a lot and historically the answer has been 'no' | 15:55 |
| arno11 | yes ofc, just out of curiosity (and to learn something) | 15:58 |
| buZz | uhuh | 16:02 |
| arno11 | buZz: still definitely slower (compared to swapfile) | 16:02 |
| buZz | right, ok | 16:03 |
| Wizzup | hm, we don't use zstd for zram do we? | 16:18 |
| Wizzup | it's lzo-rle on my d4 | 16:19 |
| Wizzup | ah we don't have this in our kernel | 16:26 |
| Wizzup | I think we want CONFIG_ZRAM_DEF_COMP_ZSTD set | 16:26 |
| uvos | Wizzup: ok | 16:58 |
| uvos | Wizzup: https://linuxreviews.org/Comparison_of_Compression_Algorithms suggests lzo is mutch faster than zstd, while being less space efficant | 17:00 |
| uvos | Wizzup: idk what is the better tradeoff | 17:01 |
| uvos | n900 is sortof short of both, space and speed | 17:01 |
| uvos | at least on d4 i find myself rarely in want of more ram but often in want of more speed so maybe there lzo is good as is? | 17:04 |
| inky | >also on pmos you could have any interface installed or no interface installed so like Wizzup sais it might be heavy or not | 17:04 |
| inky | with every interface, i usually use sxmo, it is the most lightweight. i don't turn on suspend. | 17:04 |
| inky | >could be wayland vs X | 17:05 |
| inky | in sxmo i use X. that's my usual there. | 17:05 |
| inky | >we def. worked on pp power management | 17:05 |
| inky | i guess you did something important and incredible. | 17:05 |
| inky | sincerely, without anything running, only bare sxmo which is dwm, it would be down in 3-4 hours, but it is up since the late morning and half of the battery is used. | 17:06 |
| inky | i'll try powertop. | 17:06 |
| uvos | i think mainly something is broken on your pmos install | 17:06 |
| inky | i understand, but sincerely, there's not much to be broken, that's a bare dwm with customizations. | 17:07 |
| inky | i'll try powertop. | 17:07 |
| uvos | the thing with pm is that it can break in a lot of ways | 17:08 |
| uvos | i think you should investigate it a bit and maybe file a bug with pmos against the difference | 17:08 |
| uvos | the kernel should be pretty mutch the same so the userspace you have installed must be misbehaveing somehow | 17:09 |
| inky | i'll try. | 17:09 |
| Wizzup | I thought zstd was faster and better, maybe I'm confused | 17:11 |
| Wizzup | uvos: I am not sure if the kernel is that exactly the same | 17:12 |
| uvos | Wizzup: im just going of those nummbers there, i have no futher knowlage in this area | 17:12 |
| uvos | but if those are correct then i think lzo may be the best tradeoff | 17:12 |
| Wizzup | they only measured cp time, which is compression only | 17:14 |
| Wizzup | not decompression | 17:14 |
| Wizzup | this would assume that most of the zram would then be written to more than read? | 17:14 |
| uvos | Wizzup: yeah | 17:14 |
| uvos | Wizzup: thats unlikely | 17:14 |
| uvos | Wizzup: but idk what the relavtive speed difference is during decompression | 17:15 |
| uvos | lzo is mainly very simple so i suspect its fast there too | 17:15 |
| Wizzup | I think zstd decompresses really fast | 17:15 |
| Wizzup | as you can see it's basically near gzip | 17:16 |
| uvos | well gzip uses deflate | 17:16 |
| uvos | and wikipedia claims lzo to be faster than deflate | 17:17 |
| uvos | so again idk | 17:17 |
| uvos | needs numbers i gues | 17:17 |
| uvos | Wizzup: so i found this | 17:24 |
| uvos | https://quixdb.github.io/squash-benchmark/ | 17:24 |
| uvos | it suggests zstd is mutch slower than zstd, but i dont know what level was used (or what level the kernel uses) | 17:25 |
| uvos | er zstd is mutch slower than lzo | 17:25 |
| uvos | in both compression and decompression | 17:26 |
| Wizzup | squash uses the highest levels I guess? | 17:26 |
| uvos | maybe dunno | 17:27 |
| Wizzup | yeah it doesn't list the levels, this is kind of essential | 17:27 |
| Wizzup | https://github.com/facebook/zstd/issues/3098 well | 17:28 |
| Wizzup | https://github.com/facebook/zstd?tab=readme-ov-file#benchmarks | 17:30 |
| Wizzup | ok, looks like lzo decompresses a bit faster too | 17:30 |
| uvos | so verdict? is better compression ratio worth mutch slower compression and slightly slower decompression? | 17:30 |
| Wizzup | don't really know, I could see it go either way | 17:31 |
| Wizzup | on the n900 which is really short on ram I could see zstd maybe being better | 17:31 |
| Wizzup | I don't feel like the n900's main issue is 100% cpu usage so to say | 17:32 |
| Wizzup | https://www.reddit.com/r/Fedora/comments/mzun99/new_zram_tuning_benchmarks/ | 17:35 |
| sicelo | inky: pmos does run quite a few things that can be CPU-intensive. you won't believe it, but, for example, sxmo pegs N900 cpu at 100% non-stop (well disabling conky makes things better) | 17:38 |
| Wizzup | o.O | 17:39 |
| uvos | wow thats terrible | 17:39 |
| uvos | i mean thats your problem right there xD | 17:39 |
| Wizzup | ah etc/init.d/zram comes from arm-sdk... that's why it isn't in dpkg -S | 17:42 |
| arno11 | Wizzup: btw how to check zram current algo ? | 17:47 |
| Wizzup | zramctl | 17:49 |
| arno11 | thx | 17:50 |
| arno11 | currently trying zstd | 18:16 |
| Wizzup | you can probably increase the zram size too | 18:17 |
| Wizzup | (since it compresses better) | 18:17 |
| arno11 | ah ok, yes seems to small | 18:19 |
| Wizzup | I'd say at least 30% more | 18:20 |
| Wizzup | maybe even 50% depending on your current size | 18:20 |
| arno11 | actually 175MB | 18:20 |
| Wizzup | there's also the question of whether the zram backing is your ram or a loopback device, I think it could be either | 18:20 |
| Wizzup | maybe not.. | 18:21 |
| * Wizzup googles | 18:21 | |
| Wizzup | I think that is zswap | 18:21 |
| arno11 | ok | 18:21 |
| Wizzup | or was :) | 18:21 |
| arno11 | ok :) | 18:22 |
| Wizzup | yes, so nevermind my above, zram is only in ram | 18:22 |
| Wizzup | but there is also https://wiki.archlinux.org/title/Zswap | 18:23 |
| arno11 | so lzo seems definitely a bit faster on n900 compared to zstd | 18:55 |
| arno11 | but slower than sd swapfile | 18:55 |
| arno11 | ofc i suppose it is possible to tweak zram stuff a lot | 18:56 |
| arno11 | on n900, the weak cpu seems the culprit bcs overclock improves things with both algos | 18:58 |
| Wizzup | faster how, startup times? | 18:59 |
| arno11 | yep | 18:59 |
| Wizzup | is it possible to set zstd level? | 18:59 |
| Wizzup | looks like not | 18:59 |
| arno11 | ok | 19:00 |
| arno11 | bbl | 19:04 |
| uvos | "After the page has been decompressed into the swap cache, the compressed version in the pool can be freed." | 21:46 |
| uvos | dose zswap really work like that, seams pointless to decompress the page before writeing it to disk, why not decompress it when reading it back | 21:47 |
| Wizzup | yeah, that is odd | 22:43 |
| inky | what was about that issue when whitespace rotated the screen? i experience it now with dino. every time i press whitespace with onscreen keyboard (not maemo keyboard), screen rotates. i remember diejuse had this issue. | 23:34 |
Generated by irclog2html.py 2.17.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!