Backwards compatibility
This was the main argument risen to ground MPLS-TP OAM on BFD. It's not a
better argument than grounding MPLS-TP OAM on 1731 due to its ETH deployment
plus coherence with SDH, OTN, as defended by ITU-T.
For reasons like the above, however, MPLS-TP BFD won't be backwards
compatible with previous BFD (even considering just CC/CV). They don't even
share the same codepoint.
The issue is not code point, which is the trivial part. It is reuse of the
majority of the implementation. Again, pretty basic.
I disagree. From a backward compatibility perspecting the codepoint is very
relevant. An existing implementation does not reckon the new encapsulation: the
only way to deploy this solution in a backward compatible manner is by
upgrading all the IP/MPLS nodes that have been deployed up to so far.
I think that the "reuse of the majority of the implementation" is actually the
issue. Those implementations do not have the capabilities required to be
deployed in a transport network.
Developing a non-transport capable standard will not make these boxes
transport capable but it would just make the standard useless for a transport
environment.
----Messaggio originale----
Da: david(_dot_)i(_dot_)allan(_at_)ericsson(_dot_)com
Data: 8-lug-2011 18.13
A: "Rui Costa"<RCosta(_at_)ptinovacao(_dot_)pt>, "Stewart
Bryant"<stbryant(_at_)cisco(_dot_)com>
Cc:
"erminio(_dot_)ottone_69(_at_)libero(_dot_)it"<erminio(_dot_)ottone_69(_at_)libero(_dot_)it>,
"mpls@ietf.
org"<mpls(_at_)ietf(_dot_)org>,
"ietf(_at_)ietf(_dot_)org"<ietf(_at_)ietf(_dot_)org>, "IETF-Announce"<ietf-
announce(_at_)ietf(_dot_)org>
Ogg: RE: [mpls] Last Call: <draft-ietf-mpls-tp-cc-cv-rdi-05.txt>
(Proactive Connectivity Verification, Continuity Check and Remote Defect
indication for MPLS Transport Profile) to Proposed Standard
Rui:
You wrote:
Reading something, keeping it on record, without effect in the draft and
"ignoring comments" have IMHO similar outcomes. As author of the draft you are
free to do it. These standards have a great impact
in our work, so i'm also free to write what i did.
Numerous comments did have effect on the draft and those that didn't were
either simply not actionable, were rhetorical or not constructive, and a few
had to be balanced against comments coming from the MPLS & BFD WGs. I would
translate "ingored" or "without effect" to "did not get one'e way". In the
standards process it happens.
Meanwhile as an editor of the document, I'll take the liberty of responding
to some of the points you raise...
My technical concerns regarding this draft were expressed...
...in the (ITU-T -> IETF, Feb/2011) liaison regarding it (LS281, i
believe);
...in operators' meetings' that took place during ITU-T's Feb/2011 plenary
meeting;
I and the WG don't really have access to private grumblings.
...in a comparison session that took place during that same ITU-T meeting.
Lots of other opinions were expressed as well, and they did not all agree
with you.
Some:
CC/CV
I don't understand the need for 2 types of packets: a single type allows CC;
mismatching identifiers in the same CC packets allow CV.
Besides adding complexity, we whether always activate both or potentiate
undetected mismerges.
OK, lets walk through this.
We want CV all the time so that any misconectivity can be detected, but on
the list it was expressed that the group did not want the overhead of
processing the source MEP TLV in every packet in order to achieve this. We
could carry it in every packet and have the receiver simply ignore most of
them, but then that would make the defect entry criteria compeltely random and
the exit criteria unreliable as well, not really a good design. Hence they are
separated using different ACH code points and the receiver is obliged to
process every source MEP TLV it receives. I hope this is clear.
(BTW: can't understand how we propose one ACH codepoint to CC, another for
CV, [counting other drafts, another for frame loss ...] but don't consider
assigning 1 single ACH protocol identifier codepoint >as requested by ITU-T)
Because that puts you into two protocol ID demultiplexing steps per OAM PDU
recevied to determine the intended function. Hence COSTS MORE. That is pretty
basic...
Uni P2P / P2MP
I can't see how BFD will support unidir and hence P2MP other than...
...eliminating the session "state variable" (down, init, up), aiming just
the state variables we really need, bringing us to something similar to 1731,
eventually with other bits on the wire or...
...using IP to create the reverse way, which we cannot assume per
requirements;
Will we create a complete different tool for that?
(BFD's B="bidirectional")
I would not go so far as to say "similar to 1731", there is actually a lot of
difference under the hood. As for uni-directional BFD, that is a BFD WG problem
at the moment.
Provisioning list
This is an MPLS profile/subset (and i heard) achievable through a
particular configuration. So, i expect each draft-ietf-mpls-TP-* to focus on
that profile/configuration. However, i keep seeing
references f.i. to IP encapsulations unexpected under TP's OAM.
I don't thus understand what the aim is: do we expect this in TP, are we
talking about MPLS in general?... The TP profile is never quite delimited.
Does chapter 4 contain ALL the configurable parameters list agreed to
provide in the comparison session?
It should. As for encapsulations, unless TP is in a complete island not
connected to anything (which as a network is rather useless) it will be
expected to interoperate with the rest of the MPLS architecture, and the stated
intention of tool development was that what resulted was applicable to the
broader MPLS architecture. Which means backwards compatiblity and procedures
for interoperation.
Backwards compatibility
This was the main argument risen to ground MPLS-TP OAM on BFD. It's not a
better argument than grounding MPLS-TP OAM on 1731 due to its ETH deployment
plus coherence with SDH, OTN, as defended by ITU-T.
For reasons like the above, however, MPLS-TP BFD won't be backwards
compatible with previous BFD (even considering just CC/CV). They don't even
share the same codepoint.
The issue is not code point, which is the trivial part. It is reuse of the
majority of the implementation. Again, pretty basic.
Simplicity
Whether we look to PDH, SDH, OTN or ETH, ITU-T's approach to CC is simpler:
in each flow, a standard defined nr of constant heartbeat signals (with
standard constant or provisioned period - no
auto/negotiated -) means OK. A standard defined number of misses means lost
Rx connection. An RDI, the only articulation between Rx and Tx flows,
meaningful in bidirectional applications, allows each
pear to identify Tx problems.
This OAM simplicity is the key for reliable fail finger pointing,
performance reports and protection. Also to allow scaling, more implementation
opportunities/manufacturers, which is valuable for
operators.
Well IMO there was not a lot of interest in T-MPLS until the IETF was going
to re-define it and make it compatible with IP/MPLS. So there was an industry
wide "design intent" implied here.
IMHO, between your MPLS-TP view and MPLS/IP, it becomes more and more
difficult to tell which is which.
That is because MPLS-TP is not a new techology, it is an addition to the
entire MPLS protocol suite.
Hope this helps
D
_______________________________________________
Ietf mailing list
Ietf(_at_)ietf(_dot_)org
https://www.ietf.org/mailman/listinfo/ietf