[pve-devel] [PATCH pve-network 0/8] generic sdn controller plugins + improvments

Thomas Lamprecht t.lamprecht at proxmox.com
Mon Oct 14 09:21:42 CEST 2019


On 10/11/19 1:13 PM, Alexandre DERUMIER wrote:
>>> Sorry, I did not had time to give this a proper check and it looks to start 
>>> getting serious so I did not wanted to wave this through. If then something 
>>> was off it would be much more work to fix/change it later on.. 
> 
> No problem, It was just to be sure that it wasn't lost in the mailing.
> 
> 
> 
>>> There are now quite a few plugins, (possible) involved daemons and 
>>> technologies here..
> yes. I have added a demo plugin for faucet (an sdn for ovs). It's not fully
> working, but it was more to see if the plugin implementation could match with
> differents sdn technos.
> 
> 
>>> Would you, Alexandre, mind writing up in short what different technologies for 
>>> which purpose/reason SDN has now or will gain soon and what implementations are 
>>> used or planned to? That would be great, as currently I'm missing a bit the 
>>> bigger picture, and that should be quite clear to be sure we're going in the 
>>> right direction.. 
> 
> yes sure.
> 
> Currently, I'm focusing on linux bridge, with vlan && vxlan.
> 
> The basic plugins are without sdn controller. (so layer2, no routing):
> 
> vlan
> -----
> - vlan is classic vlan, but be able to define vmbr/vnet at datacenter level.
> so we'll be able to add permission on them. (That's really missing currently, if we want to give
> access to customer to some speficic vlans).
> 
> vxlan
> -----
> vxlan:(unicast|multicast) is like vlan, but with tunneling over ip. (layer2, no routing)
> 
> 
> vxlan(bgp-evpn) is where we have a "real" sdn, with a controller. That's mean no arp flood,
> the controller (frr router here), exchange through bgp the mac/ip of the vms to differents hosts.
> And it's also possible to have anycast gateway (each host, have same ip on vmbr, which is the gateway
> of the vm).
> For external connectivity,basic bgp routing and static routing is implemented too.
> 
> (Our usecase at work, is 3 datacenters with a shared proxmox cluster/ceph where we can move vm across dc, keeping the same gateway)
> 
> That's the basic.
> (I think it's almost complete, maybe need to code cleanup, and I need to implement api permissions too)
> 

OK, thanks for the write up. I'll try to give this series a look the next days.
If something holds you up on continuing ping me and I'll try to give it more
priority.

thanks!




More information about the pve-devel mailing list