Eric, Les and all,
I concur with Eric. RTM capability is a kind of TE information that the path computation algorithm must take into account when selecting the suitable path for
the LSP that would be used for RTM purposes.
My 2c,
Sasha
Office: +972-39266302
Cell: +972-549266302
Email: Alexander.Vainshtein@xxxxxxxxxxx
From: Eric Gray [mailto:eric.gray@xxxxxxxxxxxx]
Sent: Thursday, January 19, 2017 10:32 PM
To: Les Ginsberg (ginsberg) <ginsberg@xxxxxxxxx>; Uma Chunduri <uma.chunduri@xxxxxxxxxx>; John E Drake <jdrake@xxxxxxxxxxx>; Acee Lindem (acee) <acee@xxxxxxxxx>; Alexander Vainshtein <Alexander.Vainshtein@xxxxxxxxxxx>; Greg Mirsky <gregimirsky@xxxxxxxxx>
Cc: mpls@xxxxxxxx; ietf@xxxxxxxx; isis-chairs@xxxxxxxx; gen-art@xxxxxxxx; draft-ietf-mpls-residence-time.all@xxxxxxxx; Abhay Roy (akr) <akr@xxxxxxxxx>; Robert Sparks <rjsparks@xxxxxxxxxxx>
Subject: RE: [mpls] Review of draft-ietf-mpls-residence-time-12
Les,
It’s nice to have a good discussion about technical aspects of a proposed standard, even late in the game.
But I have some difficulty is seeing how this usage is “clearly not TE information.”
This information should allow the head-end of an LSP to select a path based on desirable interface capabilities. For the purposes of accurate time
information, the best path may not also be the shortest path. Is this not exactly what traffic engineering is about?
As far as extending this argument to the ability to advertise additional link capabilities, I can easily imagine scenarios where that would make
sense. But nobody is suggesting that every LSR (or router for that matter) either must have this capability for every scenario, or that – having the capability – it would need to be turned on.
If there is an application or use-case out there that might require advertising every conceivable interface capability, then vendors will want to
take a look at the size of the opportunity represented by that and make a decision as to whether or not to support this.
--
Eric
Uma –
I readily admit S-BFD descriptors are – strictly speaking - a violation of the “non-routing info” policy – this was publicly acknowledged at the time the drafts
were written. The exception was justified on the basis that:
o IGPs are BFD clients
o The S-BFD discriminator information is unique and extremely modest in size
What we have here is a proposal to start advertising non-routing related interface attribute information. There is nothing special
or unique about RTM – it is simply one of many interface attributes which are generic in nature. Having agreed to advertise RTM in the IGPs, how would you then determine what other interface attributes should/should not be advertised by the IGPs? Take a look
at your favorite vendors box and see how many interface attributes can be configured.
It is also concerning that – when using the IGPs – we flood information which must be stored on every node even though the use case for it is limited to a much
smaller subset.
I think we can and should be smarter in this regard – and given the extensive work being done to enhance manageability I would like to see us benefit from this.
The authors of draft-ietf-mpls-residence-time clearly had some thoughts in this regard – which is why they proposed to use RFC 6823 (GENAPP) in IS-IS to advertise
the information –but the proposal in the draft is flawed in this regard. If we were to head in the “application” direction I would propose a much more generic “application” which is capable of advertising many interface attributes. However this goes back to
my original concern – whether we want to use IGPs to flood interface attribute information unrelated to routing even if it is segregated under an application. Remember that RFC 6823 stipulates that a separate instance of the protocol (a la RFC 6822) be used
for such cases.
Les
I support advertising this in IGP.
John –
For me, this raises the age-old question of when it is/is not appropriate to use IGPs for flooding information.
This is clearly not TE information – you just happen to be using this in conjunction with MPLS – but it is a generic capability. I do not see the IGPs as the
appropriate mechanism to flood generic interface capabilities. It also, as Acee has pointed out, results in flooding information to all nodes in the domain when only a few care about it.
Les
Acee,
Relying on an omniscient controller is a non-starter in general and in particular because the protocol by which it would learn each node’s RTM capabilities and
distribute them to the other nodes is undefined. Further, one of the ways by which an omniscient controller learns a node’s capabilities is by snooping the link/state database.
Yours Irrespectively,
John
From:
John E Drake <jdrake@xxxxxxxxxxx>
Date: Thursday, January 19, 2017 at 10:43 AM
To: Acee Lindem <acee@xxxxxxxxx>, Alexander Vainshtein <Alexander.Vainshtein@xxxxxxxxxxx>, Greg Mirsky <gregimirsky@xxxxxxxxx>,
"Les Ginsberg (ginsberg)" <ginsberg@xxxxxxxxx>
Cc: Robert Sparks <rjsparks@xxxxxxxxxxx>, "mpls@xxxxxxxx" <mpls@xxxxxxxx>, "gen-art@xxxxxxxx" <gen-art@xxxxxxxx>,
"draft-ietf-mpls-residence-time.all@xxxxxxxx" <draft-ietf-mpls-residence-time.all@xxxxxxxx>, "ietf@xxxxxxxx"
<ietf@xxxxxxxx>, "isis-chairs@xxxxxxxx" <isis-chairs@xxxxxxxx>, "Abhay Roy (akr)" <akr@xxxxxxxxx>
Subject: RE: [mpls] Review of draft-ietf-mpls-residence-time-12
Acee,
We discussed all of this with you over a year ago and used your guidance in adding the indication of RTM capability to OSPF.
I’m sorry but I focused mainly on the OSPF protocol aspects then and didn’t question the use case. This question came up in the IS-IS WG discussions.
Yours Irrespectively,
John
I guess what we were trying to envision the use case and whether it makes sense for all the nodes in the IGP routing domain to have this information. Would the
LSP ingress LSR only need to if the egress LSR supports RTM and it is best effort recording for transit LSRs in the path?
Additionally, if it is needed in the IGPs, should there also be a BGP-LS Link Attribute TLV proposed?
From:
Alexander Vainshtein <Alexander.Vainshtein@xxxxxxxxxxx>
Date: Thursday, January 19, 2017 at 10:15 AM
To: Greg Mirsky <gregimirsky@xxxxxxxxx>, "Les Ginsberg (ginsberg)" <ginsberg@xxxxxxxxx>
Cc: Acee Lindem <acee@xxxxxxxxx>, Robert Sparks <rjsparks@xxxxxxxxxxx>, "mpls@xxxxxxxx" <mpls@xxxxxxxx>,
"gen-art@xxxxxxxx" <gen-art@xxxxxxxx>, "draft-ietf-mpls-residence-time.all@xxxxxxxx" <draft-ietf-mpls-residence-time.all@xxxxxxxx>,
"ietf@xxxxxxxx" <ietf@xxxxxxxx>, "isis-chairs@xxxxxxxx" <isis-chairs@xxxxxxxx>, "Abhay Roy (akr)" <akr@xxxxxxxxx>
Subject: RE: [mpls] Review of draft-ietf-mpls-residence-time-12
Hi all,
I concur with Greg: from my POV an interoperable solution should not depend on an omniscient NMS client distributing information about capabilities of each node
to each other node.
Regards,
Sasha
Office: +972-39266302
Cell: +972-549266302
Email:
Alexander.Vainshtein@xxxxxxxxxxx
From: Greg Mirsky [mailto:gregimirsky@xxxxxxxxx]
Sent: Thursday, January 19, 2017 6:01 PM
To: Les Ginsberg (ginsberg) <ginsberg@xxxxxxxxx>
Cc: Acee Lindem (acee) <acee@xxxxxxxxx>; Robert Sparks <rjsparks@xxxxxxxxxxx>;
mpls@xxxxxxxx;
gen-art@xxxxxxxx;
draft-ietf-mpls-residence-time.all@xxxxxxxx; ietf@xxxxxxxx;
isis-chairs@xxxxxxxx; Abhay Roy (akr) <akr@xxxxxxxxx>
Subject: Re: [mpls] Review of draft-ietf-mpls-residence-time-12
Hi Les,
I believe that IGP extensions to advertise RTM capability are required.
On Thu, Jan 19, 2017 at 7:57 AM, Les Ginsberg (ginsberg) <ginsberg@xxxxxxxxx> wrote:
Greg –
I am having trouble understanding your response.
The question we are raising is whether we should extend the IGPs to support advertising RTM capability
– an alternative being to retrieve the capability via network management.
Saying that the IGP functionality is optional and/or wouldn’t always be advertised doesn’t really answer
the question of whether we should or should not define the IGP extensions.
Could you respond more directly to this point?
Les
Hi Acee,
the draft defines optional functionality. If an operator has no use neither for PTP's Transparent Clock, nor RTM itself as performance metric, then RTM
sub-TLV would not be included and thus it would not be flooded. Of course, it be right to reflect RTM capability through YANG data model, thus allowing SDN scenario you've described.
On Wed, Jan 18, 2017 at 2:51 PM, Acee Lindem (acee) <acee@xxxxxxxxx> wrote:
Although it is a bit late, we’ve had some discussions amongst the IS-IS and OSPF chairs and are wondering
whether the interface capability belongs in the IGPs. This will be flooded throughout the entire routing domain – is it really needed on every node or will it the RTM testing be initiated from an omniscient NMS client that would know the capabilities of each
node or easily query them using YANG?
Hi Robert,
thank you for the most expedient review and comments. I'll make changes in Section 2 per your suggestion.
On Wed, Jan 18, 2017 at 10:34 AM, Robert Sparks <rjsparks@xxxxxxxxxxx>
wrote:
The changes all look good.
I still think you should say something in the document about what "the time of packet arrival" and "transmission" means, and call out the point you made about being careful to not
introduce apparent jitter by not making those measurements consistently. (The definitions you point to in your earlier mail from G.8013 don't really help - they just say "time of packet arrival". Again, the first and last bit are likely to be several nanoseconds
apart so I think it matters. Perhaps you're saying it doesn't matter as long as each node is consistent (there will be error in the residence time measurement, but it will be constant at each node, so the sum of errors will be constant, and the clocks will
be ok?)
Please look at the new first paragraph of section 2 - there's a mix of "as case" and "in case" that should be made consistent. I suspect it would be easiest to simply say "referred
to as using a one-step clock" and "referred to as using a two-step clock" or similar.
RjS
On 1/18/17 12:03 PM, Greg Mirsky wrote:
Hi Robert,
Sasha Vainshtein came with elegant idea to address disconnection between discussion of one-step and two-step
modes that you've pointed out. We've moved Section 7 as sub-section into Section 2 now. Attached are updated diff and the proposed new version -13.
On Wed, Jan 18, 2017 at 8:13 AM, Greg Mirsky <gregimirsky@xxxxxxxxx>
wrote:
Hi Robert,
once again, thank you for your thorough review and the most detailed comments. I've prepared updated
version and would greatly appreciate if you review the changes and let us know whether your comments been addressed. Attached are diff and the new version.
On Wed, Jan 11, 2017 at 7:48 AM, Robert Sparks <rjsparks@xxxxxxxxxxx>
wrote:
Reviewer: Robert Sparks
Review result: Ready with Nits
I am the assigned Gen-ART reviewer for this draft. The General Area
Review Team (Gen-ART) reviews all IETF documents being processed
by the IESG for the IETF Chair. Please treat these comments just
like any other last call comments.
For more information, please see the FAQ at
<https://trac.ietf.org/trac/gen/wiki/GenArtfaq>.
Document: draft-ietf-mpls-residence-time-12
Reviewer: Robert Sparks
Review Date: 2017-01-10
IETF LC End Date: 2017-01-17
IESG Telechat date: 2017-02-02
Summary: Ready (with nits) for publication as a Proposed Standard
I have two primary comments. I expect both are rooted in the authors
and working group knowing what the document means instead of seeing
what
it says or doesn't say:
1) The document is loose with its use of 'packet', and where TTLs
appear when
they are discussed. It might be helpful to rephrase the text that
speaks
of RTM packets in terms of RTM messages that are encoded as G-ACh
messages and
not refer to packets unless you mean the whole encapsulated packet
with MPLS
header, ACH, and G-ACh message.
2) Since this new mechanic speaks in terms of fractional nanoseconds,
some
discussion of what trigger-point you intend people to use for taking
the
precise time of a packet's arrival or departure seems warranted. (The
first and
last bit of the whole encapsulated packet above are going to appear at
the
physical layer many nanoseconds apart at OC192 speeds if I've done the
math
right). It may be obvious to the folks discussing this, but it's not
obvious
from the document. If it's _not_ obvious and variation in technique
is
expected, then some discussion about issues that might arise from
different
implementation choices would be welcome.
The rest of these are editorial nits:
It would help to pull an overview description of the difference
between
one-step and two-step much earlier in the document. I suggest in the
overview
in section 2. Otherwise, the reader really has to jump forward and
read section
7 before section 3's 5th bullet makes any sense.
In section 3, "IANA will be asked" should be made active. Say "This
document
asks IANA to" and point to the IANA consideration section. Apply
similar
treatment to the other places where you talk about future IANA
actions.
There are several places where there are missing words (typically
articles or
prepositions). You're less likely to end up with misinterpretations
during the
RFC Editor phase if you provide them before the document gets that far
in the
process. The spots I found most disruptive were these (this is not
intended to
be exhaustive):
Section 3: "set 1 according" -> "set to 1 according"
Section 3: "the Table 19 [IEEE..." -> "Table 19 of [IEEE..."
Section 4.2: "Detailed discussion of ... modes in Section 7."
-> "Detailed discussion of ... modes appears
in Section 7."
Section 10: "most of" -> "most of all"
In Setion 3.1 at "identity of the source port", please point into the
document
that defines this identity and its representation. I suspect this is a
pointer
into a specific section in IEEE.1588.2008].
|