I just upgraded to the Asus Z790 ProArt Creator which has the 10gb nic built in. I have it connected via RJ45 at the board into a SFP+ to RJ45 adapter. On the Ubiquiti switch, it negotiates at 10GB, and in the webui main page, it says 10000. Using ethtool however, i see that the ubiquiti switch is asking for 10000, but the Asus motherboard can only advertise to 5000. I tried a different 10GB switch which has RJ45 ports and ran ethtool again and the same thing shows up, so I dont think its any fault of the switches or cables? I included the diagnostics and the screenshot below. Any help is greatly appreciated! I get occasional drop out in accessing the web ui and also over 500k read drops.

toblerone-diagnostics-20240217-0124.zip

This is the specific controller:

Ethernet controller: Aquantia Corp. AQC113CS NBase-T/IEEE 802.3bz Ethernet Controller [AQtion] (rev 03)

 

I also tried a 10GB only SFP+ and also a multi-gig SFP+ adapter and I get the same behavior.

Edited February 17, 20242 yr by Johann

48 minutes ago, Johann said:

This is the specific controller:

Ethernet controller: Aquantia Corp. AQC113CS NBase-T/IEEE 802.3bz Ethernet Controller [AQtion] (rev 03)

 

I also tried a 10GB only SFP+ and also a multi-gig SFP+ adapter and I get the same behavior.

Eth tool is showing 10Gb

 

 

 

Not sure why you are getting drops. There are comments about turning off power efficiency settings for windows, not found an option on linux yet.

usually drops at 10G mean: BAD CABLE!!!

Make sure, you have got a real 10G cable. Most of them sadly are are "raw cables" which means, the cable is fine, but the plugs are not capable of 10G.

As a last resort the speed will drop to 5 or even 2,5 G to compensate the transmission errors.

 

BTW: it has NOTHING to do with power efficence. The link speed is the same for 2.5, 5 and 10G. Just the Usage/Pause times are different. So it is not wrong or uncommon, that a switch reports a 10G link while the real used speed is lower. Make sure that Flow Control is turned on and working, else you will notice a lot of retransmissions and slowdowns.

 

BTW2: if possible, avoid twisted pair 10G completly and go safe with fiber or direct connect SFP+

 

BTW3: you HAVE a 10G connection, read your list correctly! the 10000 comes before 2500 and 5000. And this is just an offered list, the picked speed is 10000 as you can see below.

Edited February 17, 20242 yr by MAM59

The drops are caused by the macvtap interface.

I can recreate them on different machines at will. Big issue that you cannot disable the macvtap if you dont need it. The versions where just eth0 is present work without an issue.
 

 

https://forums.unraid.net/bug-reports/stable-releases/6126-macvtap-causes-consistent-package-loss-r2836/

If you wanna contribute to it^^ 


Edit: For the funsies, here is the result from my main pc, as before, completely different hardware, completely different cables.
 

Heres a comparison with an older Unraid version
 

 

 

Edited February 17, 20242 yr by Mainfrezzer

7 hours ago, SimonF said:

Eth tool is showing 10Gb

 

 

 

Not sure why you are getting drops. There are comments about turning off power efficiency settings for windows, not found an option on linux yet.

Ah yes thank you. Right after I posted I noticed that, I deleted the post cause I just didn't see it at first for some reason, I assumed it listed it in order of speed. Thank you for pointing that out.

6 hours ago, MAM59 said:

usually drops at 10G mean: BAD CABLE!!!

Make sure, you have got a real 10G cable. Most of them sadly are are "raw cables" which means, the cable is fine, but the plugs are not capable of 10G.

As a last resort the speed will drop to 5 or even 2,5 G to compensate the transmission errors.

 

BTW: it has NOTHING to do with power efficence. The link speed is the same for 2.5, 5 and 10G. Just the Usage/Pause times are different. So it is not wrong or uncommon, that a switch reports a 10G link while the real used speed is lower. Make sure that Flow Control is turned on and working, else you will notice a lot of retransmissions and slowdowns.

 

BTW2: if possible, avoid twisted pair 10G completly and go safe with fiber or direct connect SFP+

 

BTW3: you HAVE a 10G connection, read your list correctly! the 10000 comes before 2500 and 5000. And this is just an offered list, the picked speed is 10000 as you can see below.

Ah yes I didn't read the list correctly! I'll try a new cable and see if it fixes it. I did have a X540-T2 which worked well, but I did an upgrade and there was 10Gb on the motherboard and I wanted to simplify/have more PCIE slots available. I agree with you and I also prefer SFP+, much simpler and cheaper! Thank you!

6 hours ago, Mainfrezzer said:

The drops are caused by the macvtap interface.

I can recreate them on different machines at will. Big issue that you cannot disable the macvtap if you dont need it. The versions where just eth0 is present work without an issue.
 

 

 

https://forums.unraid.net/bug-reports/stable-releases/6126-macvtap-causes-consistent-package-loss-r2836/

If you wanna contribute to it^^ 


Edit: For the funsies, here is the result from my main pc, as before, completely different hardware, completely different cables.
 

 

Heres a comparison with an older Unraid version
 

 

Wow thanks for the videos! This is definitely it, I have bridging disabled because I use macvlan which I followed here https://docs.unraid.net/unraid-os/release-notes/6.12.4/, I like macvlan because it assigns a different MAC address. Is this still the recommended way to use macvlan, by disabling bridging? I have the same behavior in the RX drops as you showed in the videos.

I made a duplicate post on accident cause I thought I deleted this one, but I saw your link to your other post. Adding onto the previous question, I see that you are saying macvtap on (is this bridging off?), and macvlan on, doesn't have this issue?

 

 

2 hours ago, Johann said:

I like macvlan because it assigns a different MAC address. Is this still the recommended way to use macvlan, by disabling bridging? I have the same behavior in the RX drops as you showed in the videos.

Yes, currently the only way for docker container.

2 hours ago, Johann said:

Adding onto the previous question, I see that you are saying macvtap on (is this bridging off?), and macvlan on, doesn't have this issue?

 

Indeed, but, I currently have the ipvlan bridge enabled because macvlan bridge would cause crashes.
Given that im stuck with a fritzbox too, i picked up the lxc pluging again for any application that needs to be run on a seperate ip.


Overall you *could* just ignore the drops. The only reason i noticed them is that i got a lot of complaints about disconnects from other people after upgrading from 6.11.5 and it took me a while until i figured the issue out. So if you didnt notice anything apart from the number itself, youre good.

Look like the drops do affect plex, I changed it to eth0 with a custom ip the other day and every 10min it caused the stream to cut and rebuffer even though the playback bar indicates it had already buffered like the next 10 minutes. I switched it back to host, and it's no longer an issue.