Whoa!
Will indeed have a look on URL you posted, thanks!

RTP is perhaps the wrong reference, I should have referred the
different overlay tunnels as NVO3 "codecs", the payload inside the
RTP, the audio/video profile identifier. Then these overlay tunnel
protocols can evolve, as the DSPs evolved/evolving in the
communication systems. What do we gain here then?
- no forklift of the whole network when a new NVO3 "codec" is introduced
- use the best "codec" for your VTEP tunnel adjacencies, e.g. between
computer nodes GUE, Geneve whatever provides the best performance in
that version of the hypervisor/NIC combination. If you are connected
to the "PSTN" and your external network is based on MPLS, then go for
MPLS over UDP or your external network is LISP capable then go for
VXLAN - these gateways will most likely do their NVO3 encaps in
hardware. Hopefully one day, the hypervisors would support MPLS (two
labels) and the OCP can setup the LSP between the hypervisor and
underlay network without running LDP between the hypervisor and
underlay network. With that we could avoid the "transcoding" scenarios
described in this draft, e.g. VXLAN-MPLS-VXLAN
https://tools.ietf.org/html/draft-ietf-bess-dci-evpn-overlay-03
And having different extension versions to get translated between
those different NVO3 "codecs" will create some hurdles with upgrades,
most likely :-).

Thus I think it would be better to keep the different NVO3 codecs as
simple as possible, optimized for the NVO3 "DSPs", and have then the
NVO3 extensions in a common sister protocol, which can evolve in its
own pace; those extensions shouldn't need hardware support anyway.
That will depend on the OCP of course, but if there is some feature
that is broken in the OCP and would require extension on every packet
(and hence changes to the NVO3 "DSP"), but there are other OCPs that
can get around this - should we allow that then, choose a single OTP
that gets us into hardware dependencies on the NVO3 "DSP" design and
requires a forklift upgrades every time a new version is released?

I don't know if this kind of scenario is possible with the three
tunnel proposals, this is a hypothetical question without any research
behind - but I am worried that we are only focusing on the overlay
tunnels and not questioning what should go to via the overlay
controller protocol, how can we have several version of hardware
supported overlay tunnels, how can these evolve and should we then
decouple the VTEP control messages from the hardware dependencies and
those control messages should be able to support several overlay
tunnel protocols?

The difference between SIP messages and RTCP is that SIP messages
between a SIP UA and a SIP proxy/registrar do take a different path in
the network what the RTP messages are using - the SIP messages are off
the data path (the signalling path) and the RTCP messages are on the
data path.

Regards,
Patrick

_______________________________________________
nvo3 mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/nvo3

Reply via email to