Slow ZFS performance on LXD instances

if i take one nvme drive and us this as slog device would it improve my qrite performance?

You can disable sync with and it may get faster, but there are some down sides.

zfs set sync=disabled <name of lxd pool>

Yes, you can make a partition on your nvme for SLOG. About 50GB should be more than enough for your 2TB pool, but honestly I have some 10K HDDs that run faster then what your SSDs reporting. Something else is going on. That’s why I asked you to scrub them just in case. Like trim, scrub takes a while to run.

I’ve got some 860 evos that I am shredding and formatting to try to replicate your setup; but I am waiting on those processes. Are your drives connected to SATA ports or over USB?

ok,

i will copy all data back and the i will scrub and trim them. Before i created the pool i reset the 4 ssd disk to factory.

i have 4 x 870qvo disks. they are connected on the sata ports. (6gbs)

How much RAM do you have?

I Have,

128GB

That rules out ARC settings; at least the default ones that you posted yesterday.

can i do slog and l2arc on the same nvme device? or is this not the way to go?

Should be fine if you use separate partitions.

can i create a zpool for this and then yust 2 filesystems? or…?

Can you report zdb | grep ashift back?

You can set up a SLOG partition on two separate devices and mirror them together; yes. Same for L2ARC.

ok. i wanted slog and l2arc on yust one nvme. otherwise i need to reinstall my os.

what is the way to create 2 partitions and assign one for slog and one for l2arc?

root@esx:~# zdb | grep ashift
ashift: 9
ashift: 9
ashift: 13
ashift: 9

That’s an issue. Your ashift should be set equally, and to at least to 12 on each of those SSDs. Is that 13 by chance the ashift for the drive that you originally fed to LXD?

Run zdb and you’ll have to manually check under each of the “children” to determine if that is the case.

how can i find this?

ps i can not change the ashift? correct? so then i need to create the pool and then add it to lxd?

Run zdb and you’ll have to manually check under each of the “children” to determine if that is the case.

no the 13 is a ssd on its own. thats my download ssd

do i need to recreate the pool with ashift=12?

Is that ashift=13 on an identical drive to the ones populating the lxd zpool?

its a 2 tb. i manualy told it to use ashift=13 when i created it.

im will recreate the pool with ashift=13 i did not copy ove reverything. better now then tomorow

For 512b sectors an ashift of 9 or higher is suggested, with 4k sectors an ashift of 12 or higher is recommended. 8k sectors are recommended 13. You really can’t overset the ashift and have it impact you negatively, but you can under set it and it will hurt your performance. Yes, I believe that you will have to rebuild the zpool like before to change the ashift value.