Lots of Errors with Netonix and new 11x

[mhoppes]

Exactly. What the heck are there so many drops,
Jabbers and filtered?

Primarily the drops are of concern.

[sirhc]

Exactly. What the heck are there so many drops,
Jabbers and filtered?

Primarily the drops are of concern.

Rx Filters is sort of a trashcan of issues in the Ethernet world.

You need to use something like wireshark in the middle with a dumb "HUB" to get in the middle to nail down Rx Filters, pain in the ass. You can Google them and read horror stories tracking them down.

Can be incorrect VLAN tags and whole slew of other issues including malformed, same with all switch / ethernet manufacturers, like I said a trashcan of many types of errors.

[mhoppes]

Ok. So why would increasing the MTU from 1528 to 2000 reduce filters to almost nothing?

[sirhc]

Ok. So why would increasing the MTU from 1528 to 2000 reduce filters to almost nothing?

I have no idea Matt, I would have to put wireshark in the middle and look at the packets to know what is going on.

Obviously the packets are getting mangled or modified to some degree going through the AFX11.

Since UBNT is using a completely custom designed software radio on an FPGA and doing digital to analog conversions and vice versa on the other side of the link I would suspect it may be something running around in that area?

As I said Filters is sorta a generic trashcan error stat it is difficult to say.

We have nothing to do with determining what packets fit into what error stat as the switch core is completely self contained and we simply pole the core for the port stats. All packet forwarding is handled inside the switch core and its internal code.

Also since our switch core uses internal PHYs which are designed by the same people designing the switch core and writing the core code it is a fully integrated package.

Now our WS2 limne uses external PHYs which we can pick and chose our PHY if we want to go rogue but we are following their reference design and their external PHYs except for our 1/2.5/5G copper ports which are another vendor.

UBNT uses an external PHY with their SOC which is also separate from their FPGA.

Also ground current can mess with Ethernet errors as well as do damage to equipment. Make sure grounding is all good and service and tower grounds are bonded good. Maybe add a new service ground rod and bond to existing.

[mhoppes]

Chris,
What is the chance that 1k != 1024 in the MTU calculations in the switch? I haven't yet tried to see how low I can go.

BUT... 1528 does not appear to completely allow a 1500 MTU packet to pass with no VLANs or other encoding on it.

[lligetfa]

Testing max MTU is easy with the ping "do not fragment" switch.

[mike99]

You can also test MTU with iperf.

https://iperf.fr/iperf-doc.php