I've had a remote UnRAID server now for 2+ years using WireGuard to connect back to my OPNsense router.

 

Until the upgrade it had no issues maintaining a 200Mbps transfer rate when pulling files across an SMB share from my home UnRAID server according to my Observium history, however, after I've upgraded I'm now seeing 20Mbps normally with only a few spikes to 200Mbps.

 

I did just do an iperf3 test between the servers in both directions and it appears that WireGuard itself is working correctly as hitting 200Mbps consistently isn't an issue so could it be Rsync?

 

Any thoughts would be appreciated.

 

ETA: It's SMB causing the slowdown as switching to NFS allows the speed to go back to normal.  Did anything change with SMB in V7?

ETA: CPU usage also goes back to normal after switching to NFS.

ETA: SMB speeds are fine from my Windows machine to the UnRAID server.

Edited January 17, 20251 yr by jlficken1

48 minutes ago, jlficken1 said:

Did anything change with SMB in V7?

Probably Samba got updated, but SMB is not slower in general at least, can you retest with 6.12 to confirm if it's still faster?

I'll try to test later today as I still have 1 server that's running 6.12, however, I've update the rest to 7 over the last week since I didn't notice any issues after a couple of days.

 

Would I want the 6.12 machine to be the source, the destination, or try both directions?

ETA: For the moment I'm trying to redo all of my backups to use NFS as I transfer hundreds of GB per day so I need them to work.  You can't create a custom NFS share like you can with SMB can you?

Edited January 17, 20251 yr by jlficken1

2 minutes ago, jlficken1 said:

You can't create a custom NFS share like you can with SMB can you?

If you mean export a share with NFS, you can, you can also connect to one using UD.

Yeah I have the basic exports for each share working, however, in SMB you can go into the Samba Extra Configuration section and do this:

[rootshare]
path = /mnt/user
comment =
browseable = no
public = no
valid users = backup
writeable = no

 

I use that as a read only single path to mount via UD for all of my backup scripts to use since it has access to all of the other shares this way.  It just makes things simpler.

Ahh, you man a root share, I know that UD can do that with SMB, you don't need to use a custom settings, those can cause other issues, not sure about NFS, recommend asking there.

I hadn't noticed that before but took a look at it.

 

It appears to just add an export to /etc/exports, however, like I discovered when doing that myself the issue is that it gets removed when you edit an NFS share through the Shares tab so it's not a perfect solution.

For now I'm just going to run this every hour through User Scripts:

if ! grep -q "fsid=999" "/etc/exports"; then
  echo "Didn't Find NFS Root Export...Adding."
  echo "\"/mnt/user\" -fsid=999,async,no_subtree_check 172.16.100.60(sec=sys,ro)" >> /etc/exports
  /etc/rc.d/rc.nfsd restart
fi

 

The IP is the address of my remote UnRAID server.

Edited January 17, 20251 yr by jlficken1

Here's some version info for a couple of things.

 

V7.0.0:

CIFS - v2.45

RSYNC - 3.3.0

 

V6.12.14:

CIFS - 2.4.0

RSYNC - 3.2.7

 

I'm not seeing the same issues with speed on my own network going between either 2 V7 machines or a V7 and a V6 machine so it may still be related to WireGuard even though NFS speeds are fine over the tunnel as well as iperf3.

 

I'm waiting for my backup to finish over NFS and then I'll try connecting over WireGuard using SMB to the V6 machine and see what the speeds look like.

It's definately something to do with SMB over WireGuard in V7 as here's transfer history from the offsite machine over the last 2 weeks.

 

Top graph is when it was running 6.12.12 and the bottom graph is after the upgrade to 7.0.0 a few days ago.

 

The source machine wasn't upgrade to 7.0.0 until 1/15.

 

The bottom graph speeds jumped back to what they should be earlier this morning when I switched from SMB to NFS.

Edited January 17, 20251 yr by jlficken1

I did find a difference in the WireGuard (wireguard.ko.xz) versions between 6 and 7.

 

6.12.14:

WireGuard - 6.1.118

 

7.0.0: 

WireGuard - 6.6.68

I was able to test with V6 and V7 machines both locally and across WireGuard with SMB and here’s what I found:

 

Locally: Both versions work together with no speed issues. 
Over WireGuard: V6 can pull from V7 with no speed issues, however, V7 is slow whether pulling from either a V6 or V7 machine.

 

The WireGuard processes CPU usage seems to spike a single core when the slowdown is happening. 
 

The only thing I’ve found to work is to switch to NFS. 

Edited January 18, 20251 yr by jlficken1

Hello for NFS you can add rule

For example mine is:
 

*(rw,sec=sys,insecure,anongid=100,anonuid=99,all_squash)

 

The line automatically added to my /etc/exports become:

 

"/mnt/user/Backup" -fsid=106,async,no_subtree_check *(rw,sec=sys,insecure,anongid=100,anonuid=99,all_squash)
 

See the rule part added.

 

So i guess for read only:

*(ro,sec=sys,insecure,anongid=100,anonuid=99,all_squash)

 

Doing that the /etc/exports isn't updated or erased after changes.

 

 

Be careful because activating NFS on share makes impossible to make "exclusive access" on this SHARE.
 

Edited January 18, 20251 yr by waazaa