I'm trying to understand a performance difference between between 2 zfs filesystems on the same disk. One is filesystem made by creating a container through proxmox. e.a. /rpool/data/subvol-***-disk-* and the other is the local filesystem e.a. /rpool/ROOT/pve-1. This only happens on non sync workloads, sync workloads give roughly the same results.
I tested both from the host os, not from inside a container. It seems that the container filesystem spends more time flushing to disk?
I guess there must a logic explanation, but I can't seem te find either here on the forum or google.
container filesystem:
local filesystem:
fio config:
I tested both from the host os, not from inside a container. It seems that the container filesystem spends more time flushing to disk?
I guess there must a logic explanation, but I can't seem te find either here on the forum or google.
container filesystem:
Code:
READ: bw=246MiB/s (258MB/s), 59.8MiB/s-62.8MiB/s (62.7MB/s-65.8MB/s), io=7410MiB (7770MB), run=30104-30104msec
WRITE: bw=61.9MiB/s (64.9MB/s), 14.0MiB/s-15.8MiB/s (15.7MB/s-16.5MB/s), io=1862MiB (1953MB), run=30104-30104mseclocal filesystem:
Code:
READ: bw=524MiB/s (549MB/s), 128MiB/s-133MiB/s (135MB/s-140MB/s), io=15.3GiB (16.5GB), run=30001-30001msec
WRITE: bw=132MiB/s (138MB/s), 32.3MiB/s-33.4MiB/s (33.9MB/s-35.0MB/s), io=3949MiB (4141MB), run=30001-30001msecfio config:
Code:
rw=rw
rwmixread=80
size=500M
directory=[either the local or container filesystem]
fadvise_hint=0
blocksize=8k
direct=1
numjobs=4
nrfiles=1
runtime=30
ioengine=libaio
time_based
Last edited:
