Host bootloader documentation nitpick (and about BTRFS)

[leesteken]

The Administration Guide claims:

For EFI Systems installed with ZFS as the root filesystem systemd-boot is used, unless Secure Boot is enabled. All other deployments use the standard GRUB bootloader

However, after installing Proxmox VE (9.2 in a VM) on BTRFS, I noticed and confirmed that it also uses systemd-boot when using EFI without secure boot. Maybe someone can please correct this?

Are there any expectation about when BTRFS will be out of "technology preview"? I intend to replace install PVE with BTRFS at home to get some experience with it.

Personally, I wish Proxmox would just use GRUB (even though systemd-boot might be simpler) for all installation so nobody gets confused (or needs to ask or to explain) about the boot loader (and Proxmox does not need to test, document and maintain both).

 

[leesteken]

Are there any expectation about when BTRFS will be out of "technology preview"? I intend to replace install PVE with BTRFS at home to get some experience with it.

Some experiences with BTRFS (for non-root, all RAID1c2 coming from ZFS mirrors, running latest no-subscription with kernel 7.0) on my single node home system (running a desktop VM with GPU and USB passthrough):
The same:

• Check-summed redundant data storage, which sounds perfect for swap, is not usable for swap. I resorted to zram for both filesystems.

The good:

• Much more memory available for filesystem cache, as it does not need to be limited since the kernel can drop it any time it needs it. I don't run enough guest all the time to fill most of the memory.
• PBS garbage collection is much faster on mirrored HDDs: seconds instead of almost an hour on ZFS.

The bad:

• VMs start a little slower, and I know ZVOLs are known to be slow so this did come as a surprise. Noticeable longer load times in games (VM on SSD mirror), approaching what feels like a factor of two. Not much of a problem but the one most noticeable speed difference between ZFS and BTRFS for me.
• VMs with PCI(e) passthrough take much longer to allocate and pin all of the VM memory (probably because more memory was used as cache, see above). It's taking tens of seconds instead of more or less instant, which made me notice it.
• ZFS keeps multiple extra copies of metadata even on mirrors, but I'm not sure BTRFS does that (only the mirror duplicate?). I have multiple PBS backups in different locations just to be safe.

The ugly:

• Doing a PBS (a Debian/Proxmox container) garbage collect breaks MythTV (in a Ubuntu container) recordings (at most 7MB/s) to the same HDD mirror. Looks like MythTV's 100MB ring buffer is filled up waiting for BTRFS to write the data (during the GC). This never happened with ZFS but the GC did take much, much longer (see above).

Does anyone have tips for preventing some processes/containers/VMs hogging the BTRFS write buffers?