In addition to the routing table, the Linux networking stack provides an ARP 
cache, TTL decrement, etc.  I don't think the optimizations you mention would 
be generally useful outside of some very specialized use-cases.  We are 
developing some other optimizations that should improve overall performance.

--Justin


On Aug 15, 2010, at 7:58 PM, Romain Lenglet wrote:

> Hi,
> 
> My understanding when using a 'gre' vport is that:
> - the next hop to send a GRE packet is determined by looking up the Linux 
> routing table;
> - each GRE packet is sent through the Linux in-kernel router for forwarding.
> Is that right?
> If so, I assume that this behavior is inherited from the original gre module.
> 
> It would be nice:
> - not to look up the Linux routing table, and to rely solely on the ovsdb 
> config for the port and/or the flow definitions / actions to determine the 
> next hop for GRE packets;
> - to have all GRE packets just forwarded to another vport (of another 
> datapath or of the same datapath), like a patch vport, without going through 
> the Linux routing.
> 
> That would make it very easy to separate the encapsulated traffic and the 
> non-encapsulated traffic into different datapaths, and to concentrate all the 
> control into OpenFlow controllers without having to setup Linux routing.
> And I think it may make the gre vport implementation simpler.
> 
> What are the plans regarding the gre vport implementation?
> Is there any plan to make gre vports behave more like patch vports, to 
> simplify the control of GRE forwarding and have more control over it in the 
> controllers?
> 
> Best regards,
> --
> Romain Lenglet
> _______________________________________________
> discuss mailing list
> [email protected]
> http://openvswitch.org/mailman/listinfo/discuss_openvswitch.org


_______________________________________________
discuss mailing list
[email protected]
http://openvswitch.org/mailman/listinfo/discuss_openvswitch.org

Reply via email to