Hi David, Sorry, I thought I had replied already, but I actually didn't...
Op 09/02/2024 om 00:45 schreef David Amorín:
Hi all, We are evaluating a change in our core network architecture, currently based on VLANs with advanced network in ACS. The change we want to implement is VXLAN+BGP, using hypervisors with KVM as VTEP (Thank you @Wido Den Hollander<mailto:w.denhollander@your.online> for sharing your knowledge to the community, about this topic)
You are welcome!
Do you know what is the max number of VNIs, MACs that one hypevisor/VTE should manage in a ACS zone?
I don't know if there is a limit, but you should know that not every MAC will be learned by every hypervisor.
I understand that if we have 400 hypervisors in just one zone, with a total of 40.000 advanced networks and 120.000 MACs, each hypervisor as VTEP should know/manage the 120k MACS. Is that correct? If so, do you identify any performance issue for this hypervisor to manage these numbers?
No, it will not. The hypervisor will only learn MACs for VNIs active on the HV.
So if you have 40.000 networks you will not have a VM from every network on every HV.
Let's say you run 500 VMs on a hypervisor which are all in a different VNI. It will then only learn the MACs of those 500 VNIs, not of every VNI.
The BGP messages will be distributed, but Frr will ignore anything not needed locally.
We have some doubts about the scalability of this architecture, so any feedback / recommendations of somebody using this architecture in production will be very helpful.
I have no doubts at all. Major, really major deployments use EVPN+BGP as the backbone for their cloud environment.
Wido
Thanks
