Hi Ben, Thank you very much. Think about this case: a VM attaches to an OVS br-int through vhost(kernel), the OVS br-int interconnects with OVS br-eth through a pair of patch ports. The physical NIC eth0 is added to br-eth. A packet originated from the VM is eventually sent out form eth0. When the data path of OVS br-int forwards the packet from the TAP (associates with the VM) interface to the patch port, if the OVS patch port is implemented entirely inside OVS user space, in this case will it get performance penalty, since a switching from kernel space to user space seems to take place?
BR Juno 2015-09-15 23:10 GMT+08:00 Ben Pfaff <b...@nicira.com>: > On Tue, Sep 15, 2015 at 04:26:25PM +0800, Na Zhu wrote: > > 1, Is there any difference between ovs patch interface and kernel veth > > pair? > > Yes. OVS patch ports are implemented entirely inside OVS userspace. > That is, nothing in the kernel represents a patch port. > > > 2, Is there any difference between ovs patch interface and ovs + dpdk > > patch interface? If yes, do the ovs+dpdk optimize the patch interface; if > > no, will the patch interface be ovs performance bottleneck? > > If you're talking about the Open vSwitch code base, there's no > difference; the same code implements patch ports in each case. Patch > ports don't have a performance penalty. > > (If you're talking about the OVDK fork of OVS, I don't know.) > _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
