Skip to content
Snippets Groups Projects
New look: Off

Enable zram swap by default for some (or all?) devices

    • Closed Imported Issue created by Administrator @root

    Some devices, like the librem 5 (and I've heard the PBP too, and probably the pinephone, and others?) benefit by having zram swap, at least I see swap being used on my L5 when I enabled it with zram after ~1 day of use. I think this becomes even more useful in low-memory situations.

    $ free -m
               total        used        free      shared  buff/cache   available
Mem:            2992        1273         710         178        1007        1478
Swap:            747         182         565

    I think it might be useful to have this enabled easily for other devices, possibly automatically with postmarketos-base-zram that can supply a good default /etc/conf.d/zram-init and depend on zram-init ?

    Either it can be enabled automatically for all devices that have a UI installed (postmarketos-base-ui dependency), or it could be enabled only for specific devices by having the device package depend on it. Some kernel config is necessary (pmbootstrap!2075 (merged)), so maybe doing it per-device once the kernel has the required config is the way to go.

    Or maybe enabling it by default is a terrible idea (if you think so, explain!)

    Thoughts?

    To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information

    Child items 0

    • No child items are currently assigned. Use child items to break down this issue into smaller parts.

    Activity

    • Administrator added discussion type::feature labels · Imported

      added discussion type::feature labels

      • Administrator
        Administrator @root · Imported
        Author Owner

        zram is certainly useful on devices with less than 1 Gb RAM. I had troubles with RAM usage on my engineer-prototype 1 Gb PineTab. There it is almost vital. For other devices I'm not so sure if there is profit.

        Enabling required kernel options is cetrainly useful.

        Btw we already have "deviceinfo_swap_size_recommended" set for some devices and hadling of that in postmarketos-base

        By Alexey Min on 2021-06-24T20:33:39

        Edited by Ghost User
      • Administrator
        Administrator @root · Imported
        Author Owner

        Btw we already have "deviceinfo_swap_size_recommended" set for some devices and hadling of that in postmarketos-base

        Ya, I know, I wrote that :)

        The problem is that only configures a file-based swap. That mechanism could be changed to just install zram-init and configure it using the swap_size_recommended specified in deviceinfo...

        By clayton craft on 2021-06-24T20:33:39

      • Administrator
        Administrator @root · Imported
        Author Owner

        For other devices I'm not so sure if there is profit.

        I think these two articles do a good job of explaining why having swap is useful, even when you have a lot of RAM:

        https://haydenjames.io/linux-performance-almost-always-add-swap-space/

        https://haydenjames.io/linux-performance-almost-always-add-swap-part2-zram/

        By clayton craft on 2021-06-24T21:58:52

        Edited by Administrator
      • Administrator
        Administrator @root · Imported
        Author Owner
        • Would be useful for servers for certain.
        • Would avoid also wear leveling of SD cards and eMMC modules.

        I think it is useful for most devices :)

        By Dylan Van Assche on 2021-06-25T06:19:49

      • Please register or sign in to reply
      • Administrator
        Administrator @root · Imported
        Author Owner

        I think rising CPU usage on devices which are supposed to have crappy CPU isn't the best idea. I'd go with enabling kernel config options on all devices, adding some scripts or packages, but still go with opt-in. Maybe pmbootstrap should ask about it?

        By JuniorJPDJ on 2021-06-25T17:04:34

        Edited by Ghost User
      • Administrator
        Administrator @root · Imported
        Author Owner

        That's a good point. How about making it the default for devices that could really use it? Feedback in this thread suggests the PBP might be such a device, where paying the extra CPU overhead might be worth it: https://forum.pine64.org/showthread.php?tid=14190

        By clayton craft on 2021-06-25T17:04:34

      • Administrator
        Administrator @root · Imported
        Author Owner

        Personally I would argue that swap-on-zram is worth it on any device with a sufficiently fast CPU. It effectively gives you some extra RAM for free, and even if the device has plenty of RAM, it gives you a bit more memory to work with before the out-of-memory killer has to kick in. If anything having plenty of memory is good since then you don't end up having to put things in swap very often, so there is very little extra CPU usage.

        By Newbyte on 2021-06-25T17:56:41

      • Administrator
        Administrator @root · Imported
        Author Owner

        That's why it should be opt-in in my opinion, some workflows would need it and some workflows would just get CPU usage penalty.

        By JuniorJPDJ on 2021-06-25T18:20:49

      • Administrator
        Administrator @root · Imported
        Author Owner

        PBP has 4G RAM, I don't see it as device with constrained memory :/ I'm using PBP without any swap at all an rarely see it with more than 50% RAM usage. Maybe users workflows differ - then we can provide an option to enable zram easily, if user wants it (add that kernel option if it's missing). Maybe some short wiki page on how to do it on Alpine, whichwill look like

        sudo apk add zram-init
sudo nano /etc/conf.d/zram-init
sudo rc-update add zram-init default
sudo rc-service zram-init start

        CPU on the PBP on the other hand, is too slow for my taste.

        By Alexey Min on 2021-06-25T22:51:18

      • Please register or sign in to reply
    • Administrator
      Administrator @root · Imported
      Author Owner

      Personally I always had stability problems on devices using zram. Multiple OS' on both a RockPro64 4GB and a Raspberry Pi 3 started more or less randomly freezing and/or resetting when zram was in use.

      By Julian Groß on 2021-07-01T15:02:04

      • Administrator
        Administrator @root · Imported
        Author Owner

        I would make this opt-in, then we can enable it for devices where it's known to work well.

        I think it might be useful to have this enabled easily for other devices, possibly automatically with postmarketos-base-zram that can supply a good default /etc/conf.d/zram-init and depend on zram-init ?

        I'm thinking about a good prefix. it would be nice if we could imply in the prefix of the package that devices are supposed to depend on this (if they use it), because with postmarketos-base (+ postmarketos-base-ui) and postmarketos-config(-nftables) devices are never supposed to depend on them.

        How about:

        • devicepkg-config-zram
        • devicepkg-dev could be refactored to provide devicepkg-dev and devicepkg-config-zram subpackages
        • devicepkg could also provide device quirks in the future: #1149 (closed)

        @craftyguy: do you want to implement it? should be fairly easy, and it seems like a good improvement :thumbsup:

        By Oliver Smith on 2021-07-19T19:27:32

        Edited by Administrator
      • Administrator
        Administrator @root · Imported
        Author Owner

        I would make this opt-in, then we can enable it for devices where it's known to work well.

        If we "enable it for devices where it's known to work well", then we're not making it opt-in anymore, and it sounds a lot like my original comment about enabling it by default on some devices.

        By clayton craft on 2021-07-19T19:27:32

      • Administrator
        Administrator @root · Imported
        Author Owner

        Yes, I meant to say: not enable it on all devices (the title has "or all?"), but only on the ones where we know for sure that it gives a benefit, and it's worth the CPU overhead. So we probably mean the same thing :)

        By Oliver Smith on 2021-08-01T16:03:00

      • Please register or sign in to reply
      • Administrator
        Administrator @root · Imported
        Author Owner

        Maybe also another little note: On the original Android for my Xiaomi wt-88047 the init scripts actually have in if statement that enables zram if there is less than 1GB of ram. My guess is that that comes straight from Qualcomm. At the very least the engineers that set up this software didn't think it would be necessary on phones with 1GB or more ram.

        By Julian Groß on 2021-07-19T19:52:09

        Edited by Ghost User
      • Administrator
        Administrator @root · Imported
        Author Owner

        I think there's evidence that having swap enabled on systems with more RAM is beneficial...

        also see: #1133 (comment 611102276)

        By clayton craft on 2021-07-19T19:52:08

      • Please register or sign in to reply
    • Administrator
      Administrator @root · Imported
      Author Owner

      quoting Rob Clark "how much RAM do you have on your phone? Enabling eviction (drm/msm modparam) plus zram swap can help. I'd use zram swap that as at least as large as RAM.. (I think we use 1.5x RAM size on chromebooks, and would like to move to 2x RAM size)"

      so, please :smile: I have 8G and OOM killer said Hi to me

      Before you ask: OOM got triggered not due to regular use, but running 10 concurrent replays of AAA games.

      By David Heidelberg on 2022-06-30T22:32:21

      Edited by Administrator
    • Administrator
      Administrator @root · Imported
      Author Owner

      we decided in the pmOS core team meeting to support zswap by default on devices in main/community

      By clayton craft on 2022-07-24T06:58:34

      Edited by Ghost User
    • Administrator removed discussion label · Imported

      removed discussion label

    • Administrator added help wanted label · Imported

      added help wanted label

    • Administrator mentioned in commit 67e6a12f · Imported

      mentioned in commit 67e6a12f

      By clayton craft on 2022-12-18T19:19:30

    • Administrator mentioned in commit 14f4a479 · Imported

      mentioned in commit 14f4a479

      By clayton craft on 2022-12-18T19:23:18

    • Administrator mentioned in commit 53a2083a · Imported

      mentioned in commit 53a2083a

      By clayton craft on 2022-12-18T19:24:39

    • Administrator mentioned in commit 791fbbfd · Imported

      mentioned in commit 791fbbfd

      By clayton craft on 2022-12-18T19:26:21

    • Administrator mentioned in commit 79891c18 · Imported

      mentioned in commit 79891c18

      By clayton craft on 2022-12-18T21:28:04

    • Administrator mentioned in merge request !3752 (closed) · Imported

      mentioned in merge request !3752 (closed)

      By clayton craft on 2023-01-06T00:30:45

    • Administrator mentioned in commit 308552b3 · Imported

      mentioned in commit 308552b3

      By clayton craft on 2023-01-06T21:40:58

    • Administrator mentioned in commit 0e625e76 · Imported

      mentioned in commit 0e625e76

      By clayton craft on 2023-01-08T06:59:05

    • Administrator mentioned in commit c69e8ca0 · Imported

      mentioned in commit c69e8ca0

      By clayton craft on 2023-01-08T18:17:20

    • Administrator mentioned in commit 567eddca · Imported

      mentioned in commit 567eddca

      By clayton craft on 2023-01-08T18:24:13

    • Administrator mentioned in commit ecc30502 · Imported

      mentioned in commit ecc30502

      By clayton craft on 2023-01-08T18:30:29

    • Administrator mentioned in commit 9a08a305 · Imported

      mentioned in commit 9a08a305

      By clayton craft on 2023-01-08T19:09:31

    • Administrator mentioned in commit c5da072b · Imported

      mentioned in commit c5da072b

      By clayton craft on 2023-01-08T19:30:55

    • Administrator mentioned in commit 57154a64 · Imported

      mentioned in commit 57154a64

      By clayton craft on 2023-01-10T17:33:20

    • Administrator mentioned in commit bca632d4 · Imported

      mentioned in commit bca632d4

      By clayton craft on 2023-01-11T14:39:50

    • Administrator closed with commit bca632d4 · Imported

      closed with commit bca632d4

    Please register or sign in to reply