The MPLS WG Archive

["Christopher Poh" <frasker@hotmail.com>]

Hi experts,

I am currently pursuing my research study on traffic engineering with MPLS. 
I am interested in exploring into hierarchical routing that hides certain 
level of details about network topology when establishing a path. MPLS 
offers 2 solutions for this, namely, through explicit peering and implicit 
peering. My preference is towards implicit approach to avoid explosive 
number of remote peering. I consulted RFC 3031, but section 4.3 is quite 
brief about implicit peering and I am doubtful about its correct usage. I 
hope some experts out there can shed some light to my doubts.

1. Implicit peering is not possible for non-merging capable LSR domain.
   Is this true?


2. Given a border router, which is an egress LSR, originates a stack
   attribute for a label and distributes the information through
   implicit peering technique.

   i)  How does the LSR manage the stack attribute? (my definition
       of 'manage' in this context is distributing the stack attribute
       when required and withdrawing it when not inuse). The RFC does
       not state this clearly. I find out that the originator of the
       stack attribute cannot easily determine whether the stack
       attribute is still being used by any LSR in the domain or not
       such that it may safely withdraw the attribute to conserve memory
       and/or label usage. Furthermore the originator cannot easily
       ensure that no LSR in the domain will ever continue to use the
       stack attribute after it has issued the withdrawal of the stack
       attribute unless there is a mechanism to guarantee this. In my
       opinion when the originator broadcast or multicast the withdrawal
       message, there may be some LSRs that may not receive it due to
       link failure or packet lost or etc. So this will create problems.

   ii) Section 3.27.5 of RFC3031 states that the intermediate LSRs need
       to store information about the label and its attribute even they
       do not require it. Is this a MUST? From my point of view an
       intermediate LSR may choose not to store the information
       provided that it does not use it because the stack attribute
       will never be visible when it relays any related tagged packets
       across. Am I missing important points here?

3. From label distibution protocol standpoint, how does a label be
   distributed with its stack attribute? I can't remember seeing this
   in LDP Specification.

I would also appreciate if anyone could provide me some links to informative 
guidelines on implicit peering and its usage.

Thank you in advance.

-Chris





_________________________________________________________________
Chat with friends online, try MSN Messenger: http://messenger.msn.com