On 4/8/22 06:59, 박경원 wrote:
Hello everyone
I'm trying to setup openstack cluster with openvswitch, following the guide in link below
https://docs.openstack.org/neutron/yoga/admin/deploy-ovs-selfservice.html <https://docs.openstack.org/neutron/yoga/admin/deploy-ovs-selfservice.html>
Diagram in the link above states that compute nodes should have three interfaces(management, overlay and provider)
My question is, do I really need separated management network and overlay network? (I only have two switches)
You don't *have* to, but it's possible. The only difference in the setup is if the ml2 config file list a different IP address than the management IP, but it's ok if both are the same (it will continue to work).
It seems like overlay traffic between VMs in virtual network are encapsulated with GENEVE, will not escape to management network.
The traffic wont escape. It's just that if one VM floods the management network, your operations may become difficult. Alternatively, you can use the same wire, but with different subnets, and setup QoS in your switch, if you identified this may be a problem. That being said, with modern networking (like 2x 25 Gbits/s becoming very common), this isn't much of a problem anymore. I hope this helps, Cheers, Thomas Goirand (zigo)