On Thu, 2021-07-29 at 22:45 +0530, Gk Gk wrote:
Hi,
I learnt that routed provider networks from the below links:
https://docs.openstack.org/neutron/latest/admin/config-routed-networks.html https://www.youtube.com/watch?v=HwQFmzXdqZM
This concept, according to them solves two issues:
1. Large broadcast L2 domains with failure domains
2. If there are multiple L2 provider networks, users are confused, unable to choose a particular network.
My question is, how does routed provider networks solve the second problem mentioned above ?
it does not really what it does is force providers to create network that the normaly users just use it moves away form the idea of self service tenant network to provider or operator created networks.
If one user wants to use external network for internet, how can he choose from this one single routed provider network ?
if the deploymen is using routed networks all netwosk shoudl provide external internet connectivy as you should be routign to the out side in your datacenter infractufrue.
Or, if another user wants another provider network, how can he choose that particular network ?
they user will see a precreted list of networks that you created and marked as shared in teh env. they then choose form that list. this feature is really for the use case where the tenant is also the person that runs the cloud or has a very close relation ship like two teams in the same company wehre by they can get a view into the underlying host toplogy.
Also the same can also be achieved by creating nova aggregates or availability zones and mapping computes with particular provider network connections and ask users to use that particular nova AZ, without creating routed provider networks at all ?
you can get a simialr affect by creatign small AZ although you dont really map those to provider network the same way. although you could aligh yoru physnet definitons to yoru AZ bondaryies if you want e.g. name it physnet_az_1 and physnet_az_2 ectra. that however does nto really help since the user cannot articalte the phynet mapping sicne they wotn know you have done that. if you want to supprot self service networks with l3 routeing instead of routed provider networks whant you really want is callico. https://docs.projectcalico.org/getting-started/openstack/overview this will replace the use of ovs or linux bridge as the networkign backend with calico which will provide l3 conenctivity to teants using bgp. this will reuslt in an networkign architcxture that is similar to that used in kubernetess but with more power over things liek haveign multiple prots on the same vm and other things not supprot by a k8s pod spec and the cni interface.
Can someone clarify this for me ?
Thanks Kumar