Some weird performance issue

[Raki72]

Hello fellow Unraiders,

 

I was about to unbalance my HDDs to follow invaderone's video to convert disk1 to ZFS. By doing so, I realized that when trying to move all data from disk1 to disk2, disk3, disk4,... the performance drops from 100MB/s at the beginning to 15MB/s after 10 hours. I then stopped the unbalance / rsync task and started some investigation.

 

1. I disabled Docker, VMs and auto-start array to be on the safe side, thenrestarted my server (22 data and 2 Parity HDDs)

 

2. using top, I found this task consuming 60+% CPU:

/usr/local/bin/shfs /mnt/user -disks 8388607 -o default_permissions,allow_other,noatime -o remember=0

This tasks never seems to end.

 

3. there are also several find commands running like:

root      9382 15493  0 13:33 ?        00:00:00 /bin/timeout 7102 find /mnt/user/Exchange -noleaf -maxdepth 10

root      9383  9382  9 13:33 ?        00:00:10 find /mnt/user/Exchange -noleaf -maxdepth 10

But these find commands end after a while.

 

Any idea what  this shfs task is doing and why it consumes almost one CPU core?

 

Thanks!

unraid-diagnostics-20230716-1329.zip

Edited July 16, 2023 by Raki72

[Raki72]

Okay, I guess I identified one  issues on my own - the find process came from the Dynamix Cache DIrectories plugin.

 

But the high CPU load from shfs is still there, even after changing  /mnt/user to /mnt/cachea in the Docker and VM settings (

Any idea?

 

[itimpi]

Why do you want to convert disk1 to ZFS.   At the moment it appears that performance of disks using ZFS format in the main array is not very good.    Not saying you do not have a valid reason but thought it was worth checking it was not being done simply because it is now an option.

[itimpi]

1 hour ago, Raki72 said:

2. using top, I found this task consuming 60+% CPU:

/usr/local/bin/shfs /mnt/user -disks 8388607 -o default_permissions,allow_other,noatime -o remember=0

This tasks never seems to end.

This is the process that supports the User Share capability so you would expect it to always be running.

[Raki72]

28 minutes ago, itimpi said:

Why do you want to convert disk1 to ZFS.   At the moment it appears that performance of disks using ZFS format in the main array is not very good.    Not saying you do not have a valid reason but thought it was worth checking it was not being done simply because it is now an option.

 

I was not aware of performance issues with ZFS. Thanks for letting know. Where do you have the information from?

 

I am looking for the compression functionality in ZFS. But I am not in a hurry.

[Raki72]

31 minutes ago, itimpi said:

This is the process that supports the User Share capability so you would expect it to always be running.

Thanks a lot! Do you happen to know what wget-at-gnutls is doing? It's popping in and out in top.

[itimpi]

6 minutes ago, Raki72 said:

I was not aware of performance issues with ZFS. Thanks for letting know. Where do you have the information from?

From a number of other posts in the forum that seem to have done what looks like sensible testing.   I have seen no properly structured formal testing though but I hope that will happen some time soon.

 

If you use your server primarily for storing media then you may not get much gain from ZFS compression as most media formats are already compressed.

[Raki72]

Parts of the data is documents and documentation, which could be compressed massively. But with performance issues that won't help me. 😁

 

Thanks for your answers.