athompso at athompso.net
Tue Feb 2 18:10:03 CET 2016
As far as I know, there is still one big problem with BTRFS for Proxmox:
Running a thin-provisioned VM (i.e. QCOW2) on top of a BTRFS filesystem
on top of a regular spinning-rust HDD produces **extremely bad**
Running the identical system but on ext2/ext3/ext4 produces the expected
performance you'd get out of an HDD.
Running the BTRFS system on an SSD instead of an HDD works fine.
IIRC, it has something to with the way BTRFS journalling works, such
that whenever the QCOW2 file had to grow, there was an I/O
That testing was done in CentOS and Ubuntu, not PVE, but apparently the
problem is related to the way QCOW2 and BTRFS interact so would be
persistent across all environments.
Possibly either BTRFS or QCOW2 have fixed this by now, I don't know.
More information about the pve-user