Also cc to Jiri and Jesse, I think mandatory L3 requirement is not reasonable for tunnel port, say VxLAN or VxLAN-gpe, its intention is to L2 over L3, so L2 header is must-have, but mandatory L3 requirement removed L2.
I also think VxLAN + Eth + NSH + Original frame should be an option, at least industries have such requirements in practice. So my point is it will be great if we can support both VxLAN-gpe+ETH+NSH+Original L2 and VxLAN+ETH+NSH+Original L2, this will simplify our nsh patches upstreaming efforts and speed up merging. -----Original Message----- From: Cathy Zhang [mailto:cathy.h.zh...@huawei.com] Sent: Thursday, June 02, 2016 2:54 AM To: OpenStack Development Mailing List (not for usage questions) <openstack-dev@lists.openstack.org>; b...@ovn.org; Yang, Yi Y <yi.y.y...@intel.com> Cc: Cathy Zhang <cathy.h.zh...@huawei.com> Subject: RE: [openstack-dev] [Neutron] support of NSH in networking-SFC Looks like the work of removing the mandatory L3 requirement associated with decapsulated VxLAN-gpe packet also involves OVS kernel change, which is difficult. Furthermore, even this blocking issue is resolved and eventually OVS accepts the VLAN-gpe+NSH encapsulation, there is still another issue. Current Neutron only supports VXLAN, not VXLAN-gpe, and adopting VXLAN-gpe involves consideration of backward compatibility with existing VXLAN VTEP and VXLAN Gateway. An alternative and maybe easier/faster path could be to push a patch of " VxLAN + Eth + NSH + Original frame" into OVS kernel. This is also IETF compliant encapsulation for SFC and does not have the L3 requirement issue and Neutron VXLAN-gpe support issue. We can probably take this discussion to the OVS mailing alias. Thanks, Cathy -----Original Message----- From: Ben Pfaff [mailto:b...@ovn.org] Sent: Tuesday, May 31, 2016 9:48 PM To: OpenStack Development Mailing List (not for usage questions) Subject: Re: [openstack-dev] [Neutron] support of NSH in networking-SFC On Wed, Jun 01, 2016 at 12:08:23AM +0000, Yang, Yi Y wrote: > Ben, yes, we submitted nsh support patch set last year, but ovs > community told me we have to push kernel part into Linux kernel tree, > we're struggling to do this, but something blocked us from doing this. It's quite difficult to get patches for a new protocol into the kernel. You have my sympathy. > Recently, ovs did some changes in tunnel protocols which requires the > packet decapsulated by a tunnel must be a Ethernet packet, but Linux > kernel (net-next) tree accepted VxLAN-gpe patch set from Redhat guy > (Jiri Benc) which requires the packet decapsulated by VxLAN-gpe port > must be L3 packet but not L2 Ethernet packet, this blocked us from > progressing better. > > Simon Horman (Netronome guy) has posted a series of patches to remove > the mandatory requirement from ovs in order that the packet from a > tunnel can be any packet, but so far we didn't see they are merged. These are slowly working their way through OVS review, but these also have a prerequisite on kernel patches, so it's not easy to get them in either. > I heard ovs community looks forward to getting nsh patches merged, it > will be great if ovs guys can help progress this. I do plan to do my part in review (but much of this is kernel review, which I'm not really involved in anymore). __________________________________________________________________________ OpenStack Development Mailing List (not for usage questions) Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev __________________________________________________________________________ OpenStack Development Mailing List (not for usage questions) Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
