I googled and was told that N-way mirror should improve read performance, but test result indicate otherwise.
Is my testing method incorrect, or is there a misunderstanding?
https://openzfs.org/wiki/Features#Improve_N-way_mirror_read_performance
test result:
PVE 8.2.2
2 nvme ssd zfs mirror
fio benchmark
fio test got about 3300m/s
iostat:
If I tear down zfs pool and run same fio test on single ssd driver, also get 3500m/s
So, my expectation is that in a ZFS mirror setup, I should achieve speeds of around 6000~7000 MB/s.
Is my testing method incorrect, or is there a misunderstanding?
https://openzfs.org/wiki/Features#Improve_N-way_mirror_read_performance
test result:
PVE 8.2.2
2 nvme ssd zfs mirror
Code:
pool: data
state: ONLINE
config:
NAME STATE READ WRITE CKSUM
data ONLINE 0 0 0
mirror-0 ONLINE 0 0 0
nvme-SanDisk_Ultra_3D_NVMe_2TB_23452H800317 ONLINE 0 0 0
nvme-SanDisk_Ultra_3D_NVMe_2TB_23452H800837_1 ONLINE 0 0 0
fio benchmark
Code:
[global]
name=nvme-seq-read
time_based
ramp_time=5
runtime=90
readwrite=read
bs=256k
ioengine=libaio
direct=1
numjobs=1
iodepth=32
group_reporting=1
[nvme0]
filename=/data/benchmark.dat
size=256G
fio test got about 3300m/s
iostat:
Code:
avg-cpu: %user %nice %system %iowait %steal %idle
0.17 0.00 13.93 0.00 0.00 85.91
Device tps kB_read/s kB_wrtn/s kB_dscd/s kB_read kB_wrtn kB_dscd
nvme0n1 13095.00 1676160.00 0.00 0.00 1676160 0 0
nvme1n1 13131.00 1680768.00 0.00 0.00 1680768 0 0
If I tear down zfs pool and run same fio test on single ssd driver, also get 3500m/s
Code:
avg-cpu: %user %nice %system %iowait %steal %idle
0.00 0.00 1.09 0.00 0.00 98.91
Device tps kB_read/s kB_wrtn/s kB_dscd/s kB_read kB_wrtn kB_dscd
nvme0n1 0.00 0.00 0.00 0.00 0 0 0
nvme1n1 27622.00 3535616.00 0.00 0.00 3535616 0 0
So, my expectation is that in a ZFS mirror setup, I should achieve speeds of around 6000~7000 MB/s.
Last edited: