Zram and physical swap

[seedboxer]

looking to use Zram as priority 1 and a Physical swap as lower as Zram seems to fill up swap quickly.

Im looking at creating the following swaps at boot.

Currently, zram remains persistent but loses the /swapfile on reboot.

How can i do this as tstab gets overwritten at boot?

NAME TYPE SIZE USED PRIO
/dev/zram0 partition 640M 6.8M -2
/swapfile file 1024M 0B -3

[MichaIng]

A zram swap has always higher priority than a disk swap. The kernel is smart enough to assume this is always wanted 😎.

How can i do this as tstab gets overwritten at boot?

It does not get overwritten at boot, at least it should not 🤔. In theory if there is some I/O error or the shutdown is unclean, cutting power before changes from write buffer are synced to disk, this may cause small changes to be lost. Verify once again that both are in /etc/fstab and reboot. If indeed an entry is lost after reboot, try the same but run sync; sleep 1; reboot to force a disk sync and give it a little time to finish before rebooting. If that does not help, then there is a larger issues with your system, like general disk write issues or some sort of mount over /etc or /etc/fstab.

[seedboxer]

Thanks got it to work via

https://github.com/foundObjects/zram-swap

Although the script didn't set zram swap size correctly easy fix, and used dietpi-manager to set normal swap now have

NAME TYPE SIZE USED PRIO
/var/swap file 1000M 0B -2
/dev/zram0 partition 2G 0B 15

Works as it should and has lz4 compression which is meant to work best for SBC.

Before was using command /boot/dietpi/func/dietpi-set_swapfile

the script seems to have set things up better.

Using

vm.swappiness = 20
vm.vfs_cache_pressure = 50
vm.dirty_ratio = 60
vm.dirty_background_ratio = 2

Pi4B 4GB. Dietpi has made this little thing a beast !

[MichaIng]

has lz4 compression which is meant to work best for SBC

LZ4 is fast but has bad compression ratio. So it depends on how often the disk swap is used whether it is good or not. If ZSTD (level 1) can reduce disk swapping notably, then it should be preferred. It is nearly as fast as LZ4 but has a much better compression ratio. Of course it highly depends on use case/data etc and kernel version surely plays a role as well, but just to give a starting point for evaluation: https://linuxreviews.org/Comparison_of_Compression_Algorithms

What I take from this is that ZSTD generally is the better default/alround compression algorithm for zram/zswap and filesystem compression, and it has a wide range of compression levels where:

• 1 gives speeds close to LZ4/LZO but much better compression ratio, hence should be usually preferred when performance matters more
• 19 gives compression ratio close to LZMA/XZ/7z but much better (compression) performance, hence should be usually preferred when compression ratio matters more

Btw, why do you use zram-swap if you use a disk swap as well? In this case I'd usually use zswap instead, which does a few things a little smarter. Since zswap is a buffer for other swap spaces, it does not work on its own, so if no disk swap is wanted at all, then zram-swap is required. This is the reason why DietPi allows to create only one swap space, either as disk file or zram, but not both. If both is wanted, then we assume that zswap is the better companion to the disk swap space.

[seedboxer]

Thanks for the great info, I have the 2nd swap just as a fail safe really, i would hope its not used as the idea is to keep writes to the disk down.

A lot of people say to use a 100 Plus vm.swapiness and this doesn't work well I find for a Pi as the swap gets full quick. So I choose to have it just as a buffer. My settings seem fine now and i have changed to the compression you recommend and see how it goes.

I got this little box doing a lot all thanks to Dietpi, it has really opened my eyes to linux as a whole, thanks so much for that.

[MichaIng]

A lot of people say to use a 100 Plus vm.swapiness

Sounds unreasonable to me indeed. Any swap usage slows down the system so I'd always recommend 1 as lowest possible number. Some say that if rarely used data is swapped, there is more space for filesystem cache to speed up file reads for those cached ones. But I think the chance that some files which are read a few times during some process are needed later is often lower than the chance that data from a process which is hold in RAM are needed later. And often performance is not the only relevant aspect but disk wear as well.

As you have zram + disk swap working now, not much need to change it. But for completeness, if you want to test zswap: https://wiki.archlinux.org/title/Zswap
It's Arch Linux docs but works the same way on all Linux distros. On RPi, the kernel boot parameters can be set in /boot/cmdline.txt to make it persistent.

Both can be enabled concurrently, but I'm not sure whether this has any benefit or only downsides or whether the kernel is smart enough to use only one of both. It's discussed here, but no final answer:

• https://www.reddit.com/r/linuxquestions/comments/ju4bft/zram_zswap_or_both/
• https://unix.stackexchange.com/questions/554733/effect-of-enabling-zswap-on-zram-without-any-disk-swap/578900