On 8/6/2015 9:50 PM, Xuxiaohu wrote:
Hi co-authors,
Thanks for this update which looks much better.
I still have three questions as follows:
1. This draft assumes that NVGRE adopts the same approach to carrying
IP traffic as VXLAN. However, NVGRE has a protocol field and
therefore it could directly carry IP traffic w/o adding the fake
Ethernet header between the IP payload and the GRE header (see
https://www.ietf.org/mail-archive/web/nvo3/current/msg01520.html and
http://tools.ietf.org/html/draft-yong-l3vpn-nvgre-vxlan-encap-03#page-3.)
Hence, how about considering the simpler layer3 overlay for NVGRE?
Our reference for the NVGRE encapsulation was
draft-sridharan-virtualization-nvgre-08.txt, which is now on the RFC
Editor's queue for publication as an Informational draft in the
"Independent Submissions" stream. That RFC-to-be only discusses the use
of NVGRE to carry ethernet frames. Draft-yong-l3vpn-nvgre-vxlan-encap
discusses the use of NVGRE to carry IP packets (analogous to VXLAN-GPE),
but that draft is expired.
I have no objection to incorporating the idea of draft-yong into the
tunnel-encaps draft, if that's what the relevant WGs support, but it
would be nice to see some further discussion first.
2. It said "...Note that if none of the TLVs specifies the MPLS
tunnel type, a Label Switched Path SHOULD NOT be used." IMO, since
the LSP is signaled by label distribution protocols such as LDP,
RSVP-TE or L-BGP, why is it still necessary to specify the tunnel
encapsulation attribute to indicate whether or not an LSP should be
used.
If the tunnel encapsulation attribute is present, identifying one or
more tunnel types, the semantics are that one should try to send packets
through one of the identified tunnel types (if possible). So if MPLS is
not specified as one of the tunnel types, the use of an MPLS LSP for
transport would have to be avoided. The presence of a TLV identifying
an MPLS tunnel type would indicate that the use of an MPLS transport
tunnel is one of the permissible options for transmitting the packets.
3. Since this draft has mentioned the encapsulation of MPLS over
VXLAN/NVGRE, would you please give any clue to the scenario of this
special encapsulation?
Any IP-based overlay should be able to pass an MPLS packet with an
arbitrary label stack.
_______________________________________________
BESS mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/bess
