ZFS performace issues

Here is my setup. I have a HP DL380p G8 with an LSI 9205 HBA in IT mode. I am running Ubuntu server 18.04. I configured 4 4TB drives in an RaidZ1 with a Samaung 970 EVO as a cache drive.
zfs pool
Then use MergerFS as transport layer to easily present the pool and any other JBOD drive to my other windows PCs as a single directory via samba.
mounted drives
I used Alex Kretzschmar’s of Self Hosted and his perfect media server series as a template. https://blog.linuxserver.io/2019/07/16/perfect-media-server-2019/

My issue is that I am trying to write data to the zfs pool and I will get full speed for about 30 seconds then it will start to degrade till the transfer drops to kb/s sometimes it will come back up to full speed but it will keep going through the same cycle.

then eventually the transfer will timeout. Let me know what other information would be helpful.

No sure exactly the issue might be as I’m no zfs guru, but I would doubble check the mergerfs man pages and git hub page. As I had some issues with mergerfs and zfs write and read speeds.
there is a couple of mount options that can help with zfs and mergerfs.
Do you get the same writing issues if you share an individual drive to the windows machine and transfer?
Just that the issue might not even be zfs or mergerfs but a network bottle neck.

No I am able to transfer at full speed ~90MB/s no issue to another disk in the mergerfs group that is not in the zfs pool. I do only have gigabit network and I am only transferring locally.

Update: I destroyed the zfs pool and reformatted each disk. Then tested each individually and they all tested good. So I created a new pool and it seems to be working correctly now. My guess is that something was wrong with the pool itself.

1 Like