Hello community experts!
I have a curiosity specifically related to the EP-R8 platform. When I look online at the hardware offload capabilities it mentions that bridged interfaces are not offloaded. It mentions PPPoE interfaces are offloaded.
Quick version:
Setup:
EP-R8 latest firmware
NS-5AC-Gen2-US latest firmware
Customer router doing PPPoE
Question:
If a set of ethernet interfaces participate in a bridge-group and the PPPoE server is bound to that bridge group are the PPPoE sessions hardware offloaded even though the bridge-group traffic is not?
Longer Version:
I've thus far been trying to keep my bridge interfaces constrained to vlans on the ethernet interfaces of the EP-R8 for use of managment related traffic (not really performance heavy or needing offload performance). This leaves the native interface for all of my client traffic (and presumably all the hardware offloaded goodness that comes from no bridge).
I have come across an interesting by-design behavior with the airMax CPE devices where if you put the device in routed mode and have a management interface on a VLAN the expectation is that there is also a valid WAN IP configuration (IP/DNS/Default Gateway etc) present. With the PPPoE proxy feature and PPPoE configured from the customer router there is really no requirement for any WAN IP configuration on the device to function and still provide internet conectivity. This is desirable for me as I can leave the device in a "routed" configuration (protecting my network from all sorts of customer misconfiguration) while not needing to configure any static routes on the EP-R8 to provide connectivity to that customer (via PPPoE + PPPoE relay).
All of this portion works as expected, but when I do this UNMS and all reachability to the management IP of the device stops. What I have found is that the airMax devices, in this configuration, issue a DHCP Request without the default-gateway or domain-name-server options set. The DHCP server on the RP-R8 happily replies to the request with just the IP and subnet.
Without any routes or DNS configured as a result of no WAN configuration and only getting back an IP address on the management interface, the airMAX device is sitting there with no ability to respond to any inbound off-subnet IP traffic and no ability to resolve names. This leaves the device kind of broken from a management perspective.
After a quick and helpful discussion on the airMax forums this appears to be a by-design behavior as proper IP configuration on the WAN interface was always presumed and the MGMT vlan was intended to route through the WAN interface for DNS and any off-link routes (I've still got an ongoing sort of discussion here, but this will likely turn into a feature request as it does not appear to be a bug).
The purpose behind the bridge interface on the EP-R8 was to get efficient usage of my management IP space for all of the devices by using a single /24 for all the management interfaces across the 4 ethernet ports which will have customer radios attached. I knew that per the EP-R8 specifications that bridged traffic would not be hardware offloaded. Now that I have to move to an untagged interface configuration I have what I think are a few options:
- Assign the phiscial interfaces to a bridge group and run both mgmt DHCP and client PPPoE server on the bridge group.
- I'll go this route if I learn that PPPoE sessions established over bridge interfaces are still offloaded. (not holding my breath here)
- Assign the physical interfaces to a bridge group, and run the mgmt DHCP server on the bridge group, but keep the PPPoE server bound to the physical interface
- Not even sure this option will work. Going to do some testing on this today when I get home but I'd expect that the broadcast packets will be intercepted by the bridge before making it to the PPPoE server bound on the physical interface and this will result in no PPPoE connection for the customer device.
- Move to a unique management DHCP subnet per interface and not use bridge groups and keep PPPoE server bound to each individual interface
- This will be sure to work but then I'm running way more DHCP servers than I actually need (4) and wasting a tiny bit of address space (basically break the /24 up into 4 /26's at the cost of losing 6 usable IPs)
Looking to learn from the community on this one. I am thinking I'm just going to need to bite the bullet and split my /24 up into 4 /26's and just allocate per interface but I figured I would exhaust all my options if nothing more than for education on my part.
I am also going to be messing around with all of this in my lab and will report back my findings as well. Are there any tools to dump the offloaded sessions or connections currently from the hardware so I can tell if a PPPoE session is being offloaded without doing a "performance" based tests?
Thanks to anyone who's taking a look at this. I've been enjoying the Ubiquiti equpment and configurations and the community forums and help resources have been invaluable towards me getting this far.