[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Multi-Area TE Protocol Extensions
hi jean-philippe,
see in-line...
Jean Philippe Vasseur wrote:
>
> Hi Dimitri,
>
> At 23:25 12/12/2002 +0100, Dimitri.Papadimitriou@alcatel.be wrote:
> >your impression is too affirmative wrt to my next reply
> >(i've never said that the xro mechanisms are restricted
> >to ma-te or that this i-d should be entirely dedicated
> >to it) also the intention of opening a charter item
> >implies imho structuring the content of proposals that
> >are on the table and that may further come in, this in
> >order to achieve common mechanisms in the scope of the
> >ccamp wg - in whatever i-d btw as long as it reproduces
> >the consensus of the wg (taking a look back at gmpls
> >signalling for instance you will see that it is the
> >concatenation of many individual submissions coming
> >from various horizons)
> >
> >as said before, we have currently two methods
>
> We probably have more ... we can also add some methods consisting on
> leaking part of the topology info into the areas, ....
i referred here to the "signalling" based methods (i should
have said before, sorry) hence, i think we should scope the
discussion to these two here for the time being
> But I do agree with you that once the CCAMP charter explicitly includes the
> multi-area TE item, we will need to find a consensus and work on a single
> document.
me too
> >on the
> >table 1) constraint passing and 2) path queries, the
> >first beneficiate from the fact it is the logical
> >evolution from loose routing (the next step) the other
> >uses signalling for querying a server, as such there
> >is one fundamental question underlying the latter:
> >moving features from distributed entities to the
> >centralized ones for path computation purposes solves
> >an intermediate problem (when a unique server is
> >available) that comes back when several centralized
> >systems have to interoperate
>
> I guess you are referring to the problem of multiple central systems
> sharing the path computation requests that indeed need to be in sync. That
> of course seriously increases the level of complexity but I do not think
> this is required. In the case of multi-area TE, a determined ABR
> (statically configured or dynamically discovered) can the THE PCS serving
> the requests for the LSRs residing within a specific area. In case of
> failure, a backup PCS can be used.
it is not "required" but your last sentence clearly
indicates it will become rapidly a "recommendation"
this clearly puts limits to the current applicability
of this method - see also here below - and i think it
would be wise to start thinking about scoping this
approach (in case).
> >but also one fundamental
> >protocol aspect (which is due to the methods currently
> >proposed) using a signalling protocol like rsvp to
> >explicitly query a server is clearly questionable (*)
>
> Well, I would be more than happy to discuss this aspect of course, but not
> from a philosophical point of view ...
when i say questionable, i don't want to enter into the
arcane of protocol philosophy but the protocol consistence
& evolution - imho i think this is part of our common
responsibility as ietf'ers - this said take this as a
technical discussion, please (no bad intentions here)
> We have various options there:
> (1) try to extend an existing protocol like RSVP. Note that
> http://www.ietf.org/internet-drafts/draft-vasseur-mpls-computation-rsvp-03.txt
> requires to specify only one new message type and a single additional
> object (REQUEST-ID object). All the other objects are completely optional.
> All the LSRs have already an RSVP stack and this allows to reuse all the
> RSVP objects defined so far to pass the constraints to the PCS. So the idea is:
> - to slightly extend an existing protocol (with one new message
> type, one new object required) without impacting its scalability,
this i-d describes in details the new messages, objects and
their usage but i don't see any detail about how RSVP works
in this case ? would you please go first by yourself into
the details of what's the use of the mandatory objects in
that context and their semantic wrt respect to their current
use ? this would help me really to understand the rationales
behind this method - the issue is as follows there is no
clear explanations on why rsvp re-use is a key in enabling
a path query between a client and a server, except from the
fact it is already used but for resource reservation -
you've said "without impacting its scalability", first you
should clarify from which viewpoint since due to the above
point it is probably a method that complexifies when the
#pcs > 1 thus i would only consider that #pcs = 1 (in a
first approach) now, if you refer to the installed and
running distributed rsvp-te software on the routers and
other switches from the client side it seems the case since
you have to maintain a session (but do i have to call it a
session in this context ?) with one pcs and even if you have
several sessions (?) you would at most double or triple their
number so it is clearly limited, the problem is in the other
way around, where you have a factor n (at least), n being
the number of nodes; if each of these nodes maintains m flows
(with m>>1) well the question becomes that the dimensioning
of the number of lsp's is dependent of the capabilities of
the pcs (thus in addition, it seems you have introduced a
new issue in network dimensioning) imho scalability with
respect to the pcs (in particular wrt to rsvp usage) imho
would deserve more words in this i-d in particular when
running on the same device than the one(s) along the lsp.
> - re use all the existing objects,
this re-use is clearly what i refer to here, i clearly
understand the use of the new objects but what about the
<SESSION>, <sender_descriptor> etc. what do they represent
in this context ?
you will see that this discussion is exactly the one that
we would have in defining a new application, but where i
see another point here one start to use rsvp as p2p tunnel
to transport the info for such kind of applications
> - not require any additional protocol running on the LSRs
> (2) re invent a new protocol from scratch ...
> I personally vote for the pragmatic approach (1), let see other inputs on
> the list.
well you have really point out the two alternatives
sitting at the both end of the solution spectrum
> I won't go in the philosophical aspects ... Note that IGPs/RSVP have not
> been designed to carry Optical/SDH/SONET/...
well let me point out here they don't carry sdh, they
extend the set of bandwidth request encoding (as such
no change in the orientation of resource reservation
of this protocol)
> infos either and I think one
> could mention many other examples like this and as a matter of fact they
> perfectly make the job. I personally think that if an exiting protocol can
> be slightly extended (without impacting its scalability, ... ) and can
> solve an actual problem, this option is preferable than trying to reinvent
> a new protocol from scratch.
using your words:
1) constraint-passing would be then much more pragmatic
(since you don't have to implement one new message, only
new objects) and this without impacting existing approaches
2) if one has to implement a new message with new semantic
(outside of the scope of resource reservation even in a
generic sense) then we can legitimatelly "ask" the question
why not a dedicated protocol for it running over udp;
and this is probably the way to go for this method use
all the object encoding, errors and s.o. without being
obliged to maintain backward semantic with rsvp itself
this is what the ccamp wg did with lmp and i don't think
this is unfeasible in this case, however i would say
that this would solve only the easy part of the problem
(the issue of what happens in case of multiple pcs that
runs in parallel is still for me an open issue)
> Again I'll be more than happy to get more feed-backs on the list.
>
> >- i think this should be clarified in the kompella i-d
>
> What would you exactly want us to clarify in the i-d ?
the following paragraph says:
" Once the path computation server computes the path, if the server is
not along the path, the server is no longer responsible for maintaining
the feasibility of the path (for one thing, the server may not even know
whether the LSR established the path in the first place)."
the "if the server is not along the path" is not a
condition, this applies independently of the condition
the lsp is along the path, this should be clarified
otherwise the above paragraph requires that you have
to maintain the state of the computed path and the
actual lsp on abr's for instance! but this is this
may be what you intend to propose ? but in this case
i don't see this clearly this i-d (is my understanding
right here ?)
> >since this is independent from the fact that the server
> >is collocated on an abr or along the lsp path - back
> >to previous method using constraint passing i would
> >say constraint-passing keeps the current philosophy
> >of rsvp-te since based on the implicit assumption
> >that each hop (in particular abr in ma-te context) are
> >capable to respond to the incoming request; is the
> >latter optimal, no of course; but i would say that at
> >the end the "simplest solution to this complex problem
> >will remain", well i think that rfc 3439 phrases this
> >principle much better than i do;
> >
> >(*) understand this as a general remark people have
> >the tendency to overload/over-generalize the features
> >supported by a protocol and its applicability scope
>
> I am also hopping we will move forward on this discussion with the input of
> Service Providers on which method(s) for inter-area TE better address their
> needs.
of course, the whole ietf community is kindly invited
to provide comments and input !
thanks,
- dimitri.
> Thanks.
>
> Cheers,
>
> JP.
>
> >thanks,
> >- dimitri.
> >
> >Jean Philippe Vasseur wrote:
> > >
> > > Hi,
> > >
> > > At 10:43 11/12/2002 -0500, Cheng-Yin Lee wrote:
> > > >Dimitri, JP,
> > > >While I think an applicability section/statement is fine, I'm not sure
> > > >if a mechanism
> > > >to compute inter-area TE path belong in this ID.
> > >
> > > This was exactly my impression.
> > >
> > > JP.
> > >
> > > >We did include applications in the first discussion of this ID and in the
> > > >slides
> > > >http://www.ietf.org/proceedings/02mar/slides/ccamp-6/index.html and in
> > > >addressing
> > > >Kireeti's suggestion to include a discussion on the impact of message
> > > >size, we provided
> > > >an inter-area example for illustration purposes. However, there have also
> > > >been
> > > >suggestions in the past to not discuss applications specifically in the
> > > >draft itself.
> > > >I am fine with either way.
> > > >
> > > >thanks
> > > >cheng-yin
> > > >
> > > >Dimitri.Papadimitriou@alcatel.be wrote:
> > > >
> > > > > jp,
> > > > >
> > > > > > this ID does not propose a mechanism to compute inter-area TE path
> > > > >
> > > > > well i would say that either none of them does it, or all of
> > > > > them, more accurately route exclusions are applicable to any
> > > > > mpls network where full computation of the ero is not performed
> > > > > before the LSP is signaled, implying thus a "scope" for computed
> > > > > path thus among other applicable to multiple areas - nevertheless
> > > > > i agree that a next version of this i-d should probably include
> > > > > an applicability section covering this in a bit more details (see
> > > > > also section 4.1 for an example -
> > > > >
> > > > > this said and as for any other input or i-d provided on constraint-
> > > > > passing (as well as path query, btw) lot of work is still needed to
> > > > > achieve a decent first step in this effort;
> > > > >
> > > > > thanks,
> > > > > - dimitri.
> > > > >
> > > > > Jean Philippe Vasseur wrote:
> > > > > >
> > > > > > Hi Dimitri,
> > > > > >
> > > > > > cc'ing CCAMP - see in line,
> > > > > >
> > > > > > At 17:45 10/12/2002 +0100, Dimitri.Papadimitriou@alcatel.be wrote:
> > > > > > >jp,
> > > > > > >
> > > > > > >in fact the below list should simply be as follows for
> > > > > > >signalling-based methods: 1) constraint-passing 2) loose
> > > > > > >routing and 3) path query;
> > > > > > >
> > > > > > >and the for constraint-passing we have a several i-d's
> > > > > > >such as,
> > > > > > >
> > > > > > >http://www.ietf.org/internet-drafts/draft-lee-ccamp-rsvp-te-exclu
> > de-r
> > > > oute-01.txt
> > > > > > >
> > > > > >
> > > > > > this ID does not propose a mechanism to compute inter-area TE path
> > > > > >
> > > > > > >constraint passing may also be extended using the crank-
> > > > > > >back as listed below; the usage scenarios of these methods
> > > > > > >are described in the following:
> > > > > > >
> > > > > > >http://www.ietf.org/internet-drafts/draft-kompella-mpls-multiarea
> > -te-
> > > > 03.txt
> > > > > > >
> > > > > > >now probably it is worth spending some time in considering
> > > > > > >input coming from the tewg wrt to this effort, most of the
> > > > > > >input that you see listed above has been produced over the
> > > > > > >past year (and more) i can't imagine stepping in the protocol
> > > > > > >details without a clear consensus on what do we want to cover
> > > > > > >within the ccamp wg & w/o taking into account the tewg rec's
> > > > > > >(probably better to discuss this on the ccamp mailing list)
> > > > > >
> > > > > > Fully agree with you. Hopefully multi-area TE should be soon part
> > of the
> > > > > > CCAMP charter.
> > > > > >
> > > > > > JP.
> > > > > >
> > > > > > >thanks,
> > > > > > >- dimitri.
> > > > > > >
> > > > > > >Jean Philippe Vasseur wrote:
> > > > > > > >
> > > > > > > > Hi,
> > > > > > > >
> > > > > > > > At 13:53 10/12/2002 +0530, sachin laddha wrote:
> > > > > > > > >Hi,
> > > > > > > > >'draft-ash-multi-area-te-reqmts-01.txt' stated in section 5 as
> > > > > > > > >"Initial requirements are given here for protocol support of the
> > > > > > > multi-area
> > > > > > > > >TE methods, which include needs to support
> > > > > > > > >
> > > > > > > > >* path-computation-server (PCS) functionality
> > [kompella, lee,
> > > > > > > vasseur,
> > > > > > > > >te-qos-routing],
> > > > > > > > >* query functionality [query, vasseur],
> > > > > > > > >* crankback functionality [crankback],
> > > > > > > > >
> > > > > > > > >and, optionally,
> > > > > > > > >
> > > > > > > > >* TE feedback functionality [feedback], and
> > > > > > > > >* summary-LSA functionality [summary_lsa]."
> > > > > > > > >.
> > > > > > > > >I want to know whether cisco/juniper routers (standard ones)
> > > > supports
> > > > > > > all of
> > > > > > > > >the above extension.
> > > > > > > > >or if not,how many.......or Whether router is supposed to
> > > > support how many
> > > > > > > > >...
> > > > > > > >
> > > > > > > > I do not think any vendor supports all the method mentioned
> > > > above, but just
> > > > > > > > a subset. If you are running a network, your feed-backs on your
> > > > preferred
> > > > > > > > method is very welcome.
> > > > > > > >
> > > > > > > > JP.
> > > > > > > >
> > > > > > > > >Regards,
> > > > > > > > >---Sachin
> > > > > > >
> > > > > > >--
> > > > > > >Papadimitriou Dimitri
> > > > > > >E-mail : dimitri.papadimitriou@alcatel.be
> > > > > > >Private: http://www.rc.bel.alcatel.be/~papadimd/index.html
> > > > > > >E-mail : dpapadimitriou@psg.com
> > > > > > >Public : http://psg.com/~dpapadimitriou/
> > > > > > >Address: Fr. Wellesplein 1, B-2018 Antwerpen, Belgium
> > > > > > >Phone : Work: +32 3 2408491 - Home: +32 2 3434361
> > > > >
> > > > > --
> > > > > Papadimitriou Dimitri
> > > > > E-mail : dimitri.papadimitriou@alcatel.be
> > > > > Private: http://www.rc.bel.alcatel.be/~papadimd/index.html
> > > > > E-mail : dpapadimitriou@psg.com
> > > > > Public : http://psg.com/~dpapadimitriou/
> > > > > Address: Fr. Wellesplein 1, B-2018 Antwerpen, Belgium
> > > > > Phone : Work: +32 3 2408491 - Home: +32 2 3434361
> >
> >--
> >Papadimitriou Dimitri
> >E-mail : dimitri.papadimitriou@alcatel.be
> >Private: http://www.rc.bel.alcatel.be/~papadimd/index.html
> >E-mail : dpapadimitriou@psg.com
> >Public : http://psg.com/~dpapadimitriou/
> >Address: Fr. Wellesplein 1, B-2018 Antwerpen, Belgium
> >Phone : Work: +32 3 2408491 - Home: +32 2 3434361
--
Papadimitriou Dimitri
E-mail : dimitri.papadimitriou@alcatel.be
Private: http://www.rc.bel.alcatel.be/~papadimd/index.html
E-mail : dpapadimitriou@psg.com
Public : http://psg.com/~dpapadimitriou/
Address: Fr. Wellesplein 1, B-2018 Antwerpen, Belgium
Phone : Work: +32 3 2408491 - Home: +32 2 3434361