Hello,
First of all, thank you very much @spirit for bringing the SDN feature to Proxmox; it will be a truly great addition!
However, we looked at the EVPN setup, and it doesn't work for us in its current state: the GUI or /etc/pve config files don't allow me to generate a configuration that works for what I want to do. No big deal, we can still implement our SDN with hooks and stuff, but I'd like to offer a couple of suggestions to make the feature more useful in complex environments like ours. For a bit of context, this is for a large infrastructure that we are designing, multi-PVE-cluster, mixed physical and virtual, with a hardware EVPN fabric; i.e. we need to be able to put PVE VMs as well as physical switch ports on our software-defined VLANs.
1- Template-based configuration
I think it would be helpful to make the FRR and ifupdown2 configuration templates available to the user: instead of a parameter form, the user would be able to choose between a few "common" configuration templates, or specify his own (e.g. as PVE snippets). In our case it could allow us to properly interop with our hardware EVPN fabric.
2- On-demand configuration
Another issue we have with the current version of the feature w.r.t our needs, is that when we create a VRF and subnet, it will bring up the VRF and interfaces on every node of the PVE cluster. But for us, one of the main benefits of going L3 is for scaling, by not needing every VLAN to be present at every hypervisor (but still being able to bring any VLAN to any hypervisor). It would be much more beneficial if the SDN subsystem had the intelligence to only bring up networks that are used on active VMs or CTs on each node.
Thanks for your consideration
Again, these are no showstoppers for us, it just means we don't get to use the built-in feature to do the job; but I still wanted to make the suggestion
Cheers,
Arthur
First of all, thank you very much @spirit for bringing the SDN feature to Proxmox; it will be a truly great addition!
However, we looked at the EVPN setup, and it doesn't work for us in its current state: the GUI or /etc/pve config files don't allow me to generate a configuration that works for what I want to do. No big deal, we can still implement our SDN with hooks and stuff, but I'd like to offer a couple of suggestions to make the feature more useful in complex environments like ours. For a bit of context, this is for a large infrastructure that we are designing, multi-PVE-cluster, mixed physical and virtual, with a hardware EVPN fabric; i.e. we need to be able to put PVE VMs as well as physical switch ports on our software-defined VLANs.
1- Template-based configuration
I think it would be helpful to make the FRR and ifupdown2 configuration templates available to the user: instead of a parameter form, the user would be able to choose between a few "common" configuration templates, or specify his own (e.g. as PVE snippets). In our case it could allow us to properly interop with our hardware EVPN fabric.
2- On-demand configuration
Another issue we have with the current version of the feature w.r.t our needs, is that when we create a VRF and subnet, it will bring up the VRF and interfaces on every node of the PVE cluster. But for us, one of the main benefits of going L3 is for scaling, by not needing every VLAN to be present at every hypervisor (but still being able to bring any VLAN to any hypervisor). It would be much more beneficial if the SDN subsystem had the intelligence to only bring up networks that are used on active VMs or CTs on each node.
Thanks for your consideration
Again, these are no showstoppers for us, it just means we don't get to use the built-in feature to do the job; but I still wanted to make the suggestion Cheers,
Arthur