The MPLS WG Archive

Cell Relay Retreat>MPLS WG Archive>month:2002-Mar> msg00268



[Date Prev][Date Next][Thread Prev][Thread Next]  
  [Date Index][Thread Index][Author Index][Subject Index]

LDP Label WIthdraw doubt !!

  • From: Eric Gray <eric.gray@sandburst.com>
  • Date: Fri, 29 Mar 2002 11:47:08 -0500
  • Cc: "Vijayanand C - CTD, Chennai." <vijayc@ctd.hcltech.com>, Shahram Davari <Shahram_Davari@pmc-sierra.com>, Abhijit Gadgil <gabhijit@ee.iitb.ac.in>, mpls@UU.NET

Eric Gray wrote:

> You're _all_ right, starting with Shahram.
>
> The intention of step LWd.5 is to say that you don't ask for
> a new label just because the current one has been withdrawn
> if you're using ordered control _and_ you're supposed to be
> waiting for a request from an upstream peer to drive your
> downstream request.
>
> It might be possible to add a note to clarify this, but it's not
> clear what the note should say.  It is very implementation
> specific - i.e. - one implementation might simply immediately
> request a new label if it is the ingress (simply because it needs
> one in order to _be_ an ingress :-)) while another may wait (as
> Abhijit and Vijay suggested) until it has an indication that its
> peer is likely to have a new label to give.
>
> Remembering that - in basic LDP - there is no hard requirement
> that an entire network must conform to either a specific label
> advertisement mode (DoD or DU) or control mode (ordered or
> downstream on demand), it is probably slightly more correct for 

This was supposed to say "(ordered or independent)".  Oops...

>
> the ingress implementation (A in the example) to immediately
> request a new Label from its downstream peer relative to the
> given FEC (B in the example).  Otherwise, A makes assumptions
> about B's behavior that may not be correct. 
>
> If A requests a label from B and B is not yet prepared to give it
> one, it could propagate the request (ordered control) or it could
> return a Notification message with no route.  If it does the latter,
> however, there will be a synchronization problem if has not also
> removed or withdrawn its route advertisement for the FEC.  If
> it has removed its route advertisement, then there will be a new
> FEC recognition event at some point in the future.
>
> -- 
> Eric Gray
>
>
> Vijayanand C - CTD, Chennai. wrote:
>
>> I also agree with abhijit that the problem can be addressed at the 
>> routing
>> protocol level. If a link goes down the routing protocol would generally
>> give a next hop change to the LDP module to trigger a request along 
>> the new
>> path. In other words, by the time LDP detects it routing protocol 
>> would have
>> sensed it and found and alternate path.If that were not the case, ie, if
>> alternate next hop is not immediatly available, the routing entity in
>> upstream(A in this case) would also be aware of that the dest is 
>> unreachable
>> and would have  informed the LDP entity about it. Subsequently when 
>> next hop
>> becomes available at B, routing protocol at A would intimate LDP at 
>> A. The
>> case described where the ingress has to re-request would seldom arise 
>> with a
>> good routing protocol implementation.
>>
>> Please correct me if I am wrong
>>
>> Regards,
>> Vijay
>>
>> -----Original Message-----
>> From: Abhijit Gadgil [mailto:gabhijit@ee.iitb.ac.in]
>> Sent: Friday, March 29, 2002 5:56 PM
>> To: Manoj Dutta
>> Cc: mpls@UU.NET
>> Subject: Re: LDP Label WIthdraw doubt !!
>>
>>
>>
>> Manoj,
>>
>> Consider here the events that would lead to such a behavior. One cannot
>> neglect the role played by routing protocol here. The first event would
>> cause B to withdraw and A to release label(s). However the next event 
>> when
>> B sees D as next hop for some fec F, would in turn cause the routing
>> protocol to indicate the reachability for fec F through D. This event
>> would generate an internal trigger at LSR A for LDP to issue request
>> (implementation dependent). I dont think it is required for A to do
>> something different becuase it is ingress LSR.
>>
>> LDP working alone without any routing protocol cannot obviously catch 
>> the
>> dynamic effects.
>>
>> -abhijit.
>>
>> Manoj Dutta wrote :
>>
>>> Hi,
>>>
>>> I have a doubt regarding label wihdraw procedures described in RFC 3036
>>> section A.1.5.
>>> Consider the following topology :-
>>>
>>>
>>>   A-------B---------C--------D
>>>           |                  |
>>>            -----D------------
>>>
>>>
>>> Now, lets suppose that for a given FEC F,
>>>
>>> a) C is B's nexthop for FEC F
>>> b) B is A's nexthop for FEC F
>>>
>>> All the routers are using Downstream on demand with ordered control. 
>>> So,
>>> A send request to B (for FEC F) and gets a label
>>> L (assuming B has a label from C and so on..). Now, for some reason the
>>> link between B and C goes down and consequently
>>> B withdraws label L from A. So, now A has no label for FEC F. After 
>>> some
>>> time, B chooses D as the nexthop for FEC F and
>>> gets a label from D subsequently. But, A's nexthop still points to B 
>>> for
>>> FEC F and so A never sends a request to B to get a label
>>> for FEC F. This behaviour is due to LWd.5 in section A.1.5 in the RFC.
>>> Is there a reason for this check. So, basically A would send
>>> a request to B if it were in DOD Independent control. According to me,
>>> if two LDP peers are using DOD then, in case of a label withdrawal the
>>> upstream should always send a request to nexthop for theFEC in 
>>> question.
>>>
>>> Does it make sense ??
>>>
>>>
>>> Thanks in advance
>>> Manoj
>>>
>>
>
>


-- 
--
Eric Gray (mailto:eric.gray@sandburst.com)
http://www.mindspring.com/~ewgray