Raspberry Pi OS Now Defaults To 512MB Swap, Updates Labwc Compositor

[phoronix]

Phoronix: Raspberry Pi OS Now Defaults To 512MB Swap, Updates Labwc Compositor

With last month's Raspberry Pi OS update they now default to Wayland on all Raspberry Pi models alongside various other operating system improvements. Out today is the latest iteration of the Debian-based Raspberry Pi OS with software updates and other changes...

https://www.phoronix.com/news/Raspberry-Pi-OS-Nov-2024

[fitzie]

did a fresh install of fedora, and my swap is zram now. not sure if there's any reason for physically based swap that would only wear down my boot ssd these days.

[Quackdoc]

did a fresh install of fedora, and my swap is zram now. not sure if there's any reason for physically based swap that would only wear down my boot ssd these days.

zram doesn't prevent hardlocks when it gets full, it just prevents it from getting full as fast. Usually hopefully fast enough that it wont need to hardlock in the first place, but if you are on a 2gb device or less, I would still have a backing real swap partition

[edxposed]

swapfile is almost useless in 2024, even zram_writeback is much better than traditional swapfile

[Linuxhippy]

swapfile is almost useless in 2024, even zram_writeback is much better than traditional swapfile

It depends. Swap as a means to run more processes than would fit in RAM is really dead, yes.

But swap as an almost-grave for all those memory pages that get allocated but almost never read, actually makes a lot of sense. Keep in mind, every RAM page freed can be automatically used by the page cache.

[edxposed]

It depends. Swap as a means to run more processes than would fit in RAM is really dead, yes.

But swap as an almost-grave for all those memory pages that get allocated but almost never read, actually makes a lot of sense. Keep in mind, every RAM page freed can be automatically used by the page cache.

For such use cases, either zswap or zram (with backing_dev) is a much more reasonable option, either for performance or disk lifetime considerations.

[Quackdoc]

For such use cases, either zswap or zram (with backing_dev) is a much more reasonable option, either for performance or disk lifetime considerations.

does zram with backing_dev write compressed pages to disk, or does it uncompress them first?

[polarathene]

does zram with backing_dev write compressed pages to disk, or does it uncompress them first?

It's used for incompressible pages as well AFAIK, so I'd assume the pages are uncompressed when written. Similar to zswap IIRC with it's in-memory compressed pool vs swap devices on disk.

[ahrs]

swapfile is almost useless in 2024, even zram_writeback is much better than traditional swapfile

Swapfile is easier to setup though if you don't have any devices to use as a writeback device. I setup ZRAM to use a loopback device as the writeback device. I don't know if this actually does anything or not.

$ cat /sys/block/zram0/backing_dev
/dev/loop0p1
$ losetup
NAME       SIZELIMIT OFFSET AUTOCLEAR RO BACK-FILE               DIO LOG-SEC
/dev/loop0         0      0         0  0 /dev/root/swap/swap.img   0     512
$ lsblk
NAME        MAJ:MIN RM   SIZE RO TYPE MOUNTPOINTS
loop0         7:0    0    16G  0 loop                                       └─loop0p1   259:11   0    16G  0 part

I don't know how to figure out if this is actually being used or not.

[fitzie]

zram doesn't prevent hardlocks when it gets full, it just prevents it from getting full as fast. Usually hopefully fast enough that it wont need to hardlock in the first place, but if you are on a 2gb device or less, I would still have a backing real swap partition

new box has four ddr5 slots, so I got four 48GB eudimms. having 100gb of buffer cache has made nautilus tolerable.