On Dec 10, 2013, at 7:52 AM, Wang, Baoyuan <baoyuan.w...@tekcomms.com> wrote:

> Simply reading some openstack manual, I found that Openstack Neutron would 
> need and/or configure several OVS bridges within every compute Nodes and 
> Neutron networking node. From OVS point of view,  the multiple OVS bridges 
> will be multiple Datapath inside one OpenvSwitch.   the data processing will 
> have to work on every flow tables of every related DP in sequence.  For 
> example, for the following path,
>      VM1 -> ( vnet0 on ovs_br1 -> veth_pair_A ) -> ( veth_pair_B on ovs_br2 
> -> eth1)
>  
> For traffic coming out of VM1, I think OVS would have to first work on the 
> flow table of ovs_br1; then output to port veth_pair_A.  Then for some future 
> loop, OVS will have to repeat it logic on the flow table of ovs_br2, etc. 
>  
> First of all, please correct me if my understanding of OVS is wrong.   If my 
> understanding is correct, did anyone evaluate the performance effect out of 
> the multiple tier OVS bridges within every hosting machines?  Any  
> enhancement to make it more efficient ?

I don't know much about how Neutron wires up topologies.  However, if it's as 
simple as you've indicated above, you may want to look at using patch ports 
instead of veths.  They connect two OVS bridges, but in the kernel datapath, 
the performance will be as if there was only one.  This optimization was added 
in OVS 1.10.

--Justin


_______________________________________________
discuss mailing list
discuss@openvswitch.org
http://openvswitch.org/mailman/listinfo/discuss

Reply via email to