The MPLS WG Archive

[Abhijit Gadgil <gabhijit@ee.iitb.ac.in>]

Eric,

If I understand what u have stated correctly, R2 would send mapping for
p.q.r.0/24 only when R1 _requests_ so. Will it not be best on R2s part to
send NO_ROUTE notification in that case?

But I am still not quite clear why R2 won't send mapping for p.q.0.0/16
unsolicitedly as u have pointed out?

regards,

-abhijit

On Fri, 8 Jun 2001, Eric Gray wrote:

 >Abhijit,
 >
 >    Sorry, forgot to 'lint' my email message.
 >
 >    I believe your premise may be incorrect below: R2 is likely not to
 >distribute a label mapping for p.q.0.0/16 since it may know what the
 >routing topology looks like from R1's perspective.  Even if R2 does
 >distribute this mapping, R1 is free to release the one it can't use and
 >ask for a different one it can use (with p.q.r.0/24).
 >
 >    The use of the word 'exact' in section 3.5.8 seems problematic.  I
 >believe the word 'unambiguous' would have been better.  If you look
 >at the procedures for handling a label request, the test (LRq.2, page
 >97) only checks if there is a next hop for the requested FEC, and there
 >is.  Okay, so you have to work a little harder to figure that out.  So?
 >
 >    Note that this would not be the case if R1 and R2 had these route
 >entries in reverse (i.e. R1 shows R2 as the next hop for p.q.0.0/16 and
 >R2 shows R3 as the next hop for p.q.r.0/24).  In that case, R2 could not
 >find an unambiguous next hop for the FEC p.q.0.0/16 requested by R1.
 >
 >--
 >Eric Gray
 >
 >You wrote:
 >
 >> Eric,
 >>
 >> If R2 is distributing bindings unsolicitedly, it will always distribute
 >> bindings for p.q.0.0/16 and won't distribute for p.q.r.0/24 as it does not
 >> have entry for that FEC in its routing table.
 >>
 >> And R2 can send Mapping for label request p.q.r.0/24 when it receives
 >> explicitely such request and in that case it MUST send a No Route
 >> Notification.(according to rfc3036 section 3.5.8).. I accept it is bit
 >> harsh on R2's part in doing so, but this is what the RFC says....