Removing veth pair between linux bridge and OVS bridge

asked 2015-06-30 09:38:44 -0500

tab87vn gravatar image

Hi,

I'm doing some experiments with OpenStack Juno. I notice that the network throughput between VM instances is horribly lower than that of Host to Host communication.

Is it true that the veth pair qvb/qvo between linux bridge, to which a VM is attached, and OVS bridge (br-int) is the culprit? What if we remove these veth pair and connect linux bridge to OVS bridge via OVS internal port? Does it cause any problem with the Security Group Policy implementation?

Thanks a lot!

edit retag flag offensive close merge delete

Comments

Are you using vhost_net?

darragh-oreilly gravatar imagedarragh-oreilly ( 2015-06-30 13:05:08 -0500 )edit

i don't think so, at least from the installation packages that I use for Juno.

tab87vn gravatar imagetab87vn ( 2015-06-30 14:34:45 -0500 )edit
1

I found it made a huge difference. Just ran modprobe vhost_neton the compute node, then started a new VM. http://www.linux-kvm.com/content/how-... . Anyway security groups won't work without the inline linuxbridge which is connected to ovs br-int with a veth

darragh-oreilly gravatar imagedarragh-oreilly ( 2015-07-01 00:55:40 -0500 )edit

@darragh-oreilly thanks, it indeed works. VM-to-VM throughput reaches 2.3 Gbps (out of 10Gb bw) now. But it seems that it pushes external network traffic through Tenant data network as well.

tab87vn gravatar imagetab87vn ( 2015-07-21 11:33:01 -0500 )edit