Guest performance

The best results for Windows guest:

• HDD (no SSD)
  • probably CEPH better handles bunch of ordered requests (like for rotational drives)
• no DISCARD
  • very poor performance with DISCARD

Ext4 is extremely slow on RBD block device.

SSD emulation enabled, discard support enabled, SCSI single controller, IO thread.

Guest Kernel 6.0.

BTRFS detects SSD and use: single data, duplicated system and metadata.

Simple and fast test show huge difference between ext4 and btrfs.

fio --name=fiotest --directory=./2ext4 --size=256m --rw=randrw --bs=4K --direct=1 --ioengine=libaio --iodepth=64 --numjobs=4 --runtime=10s --time_based=1

Ext4 with unsafe options (barrier=0,data=writeback)

 READ: bw=981KiB/s (1005kB/s), 234KiB/s-254KiB/s (240kB/s-260kB/s), io=10.5MiB (11.0MB), run=10981-10991msec
WRITE: bw=1030KiB/s (1055kB/s), 251KiB/s-266KiB/s (257kB/s-273kB/s), io=11.1MiB (11.6MB), run=10981-10991msec

Ext4 with default options:

 READ: bw=997KiB/s (1021kB/s), 242KiB/s-267KiB/s (248kB/s-274kB/s), io=10.6MiB (11.1MB), run=10583-10840msec
WRITE: bw=1048KiB/s (1073kB/s), 259KiB/s-271KiB/s (266kB/s-277kB/s), io=11.1MiB (11.6MB), run=10583-10840msec

BTRFS with default options:

 READ: bw=3281KiB/s (3359kB/s), 806KiB/s-855KiB/s (826kB/s-875kB/s), io=33.6MiB (35.3MB), run=10496-10497msec
WRITE: bw=3431KiB/s (3513kB/s), 850KiB/s-873KiB/s (870kB/s-894kB/s), io=35.2MiB (36.9MB), run=10496-10497msec

BTRFS with nodatacow:

 READ: bw=2444KiB/s (2502kB/s), 604KiB/s-619KiB/s (618kB/s-634kB/s), io=25.1MiB (26.3MB), run=10469-10501msec
WRITE: bw=2566KiB/s (2628kB/s), 629KiB/s-652KiB/s (644kB/s-668kB/s), io=26.3MiB (27.6MB), run=10469-10501msec

 READ: bw=3109KiB/s (3184kB/s), 770KiB/s-796KiB/s (789kB/s-815kB/s), io=30.0MiB (32.5MB), run=10193-10195msec
WRITE: bw=3253KiB/s (3331kB/s), 786KiB/s-847KiB/s (805kB/s-867kB/s), io=32.4MiB (33.0MB), run=10193-10195msec

BTRFS converted to SINGLE with command

btrfs balance start --force -sconvert=single -mconvert=single ./1btrfs/

 READ: bw=3126KiB/s (3201kB/s), 755KiB/s-815KiB/s (773kB/s-835kB/s), io=31.4MiB (32.9MB), run=10283-10286msec
WRITE: bw=3267KiB/s (3345kB/s), 793KiB/s-864KiB/s (812kB/s-884kB/s), io=32.8MiB (34.4MB), run=10283-10286msec

 READ: bw=3861KiB/s (3953kB/s), 941KiB/s-984KiB/s (964kB/s-1008kB/s), io=39.2MiB (41.1MB), run=10341-10400msec
WRITE: bw=3997KiB/s (4093kB/s), 971KiB/s-1034KiB/s (995kB/s-1059kB/s), io=40.6MiB (42.6MB), run=10341-10400msec

+ nodatacow:

 READ: bw=3911KiB/s (4004kB/s), 959KiB/s-1004KiB/s (982kB/s-1028kB/s), io=39.3MiB (41.2MB), run=10290-10295msec
WRITE: bw=4049KiB/s (4147kB/s), 993KiB/s-1039KiB/s (1016kB/s-1064kB/s), io=40.7MiB (42.7MB), run=10290-10295msec

 READ: bw=3879KiB/s (3972kB/s), 941KiB/s-1001KiB/s (963kB/s-1025kB/s), io=38.3MiB (40.2MB), run=10110-10114msec
WRITE: bw=4028KiB/s (4125kB/s), 976KiB/s-1041KiB/s (999kB/s-1066kB/s), io=39.8MiB (41.7MB), run=10110-10114msec