How to architect systems to improve Network node scalability?

asked 2014-07-11 09:30:32 -0600

kevin gravatar image

updated 2014-07-11 14:03:18 -0600

smaffulli gravatar image

Hi all,i am still using Folsom release with nova-network in production and planning to upgrade to Grizzly release,before that can some one please clarify this.


If i choose NAT, then will the private IP of VM be routed to virtual router(gateway) that situates in Network node and there it do a NAT

if this the case can VM traffic reaching neutron server can be distributed to multiple neutron servers?

Case 2

does the virtual router situate in Hypervisor itself in different name space and do the NAT ing in Hypervisor itself this is the case with nova-network except that there is no namespace

As per below diagram it's not showing any public IP on compute nodes, this is confusing me (As NAT on hypervisor will not work in this case), so i am assuming Case 1 is the grizzly way of NAT and all the traffic goes through network node.

image description

Can this network node be distributed in some way as i believe a single server can not handle traffic of all the VMs coming from 100's of compute node, i am sure there should be a way since the new neutron model is advanced version of nova-network.

Can i completely avoid this virtual router concept if i use two flat networks,one with public and other with private?

edit retag flag offensive close merge delete