The MPLS WG Archive

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

On Mon, 2 Jul 2001, Manoj Dutta wrote:

 >Hi,
 >
 >I have the following doubts in RFC 3036.
 >
 >1) In LMp.6 and LMP.7, what happens if the label being deleted maps to
 >one or more upstream labels ?  If those labels are not withdrawn, then
 >the LSR in question would end up  discarding all the incoming packets
 >bearing those labels.

Sending withdraw message is not required here, as we _do have_ outgoing
label for all those incoming upstream labels. What basically is being
tried to achieve here is to stop an attempt to create multiple bindings.
The newly learned label is perhaps _better_ so use it instead of the old
label. Why would u send withdraw. If u send withdraw u will also have to
send another mapping perhaps unsolicitedly.

If the attributes of newly received label are different than those of
previously received one, that will be taken care of in subsequent
processing. (LMp. 17 loop)

 >
 >2) In LMp.23 (Note 9), only path vector attribute is considered.  Is
 >there a reason for not considering hop count attribute.

It does not explicitely prevents from checking Hop Count. On the other
hand hop-count should be checked. Note 9 is illustrating on what to do
with path vectors.

 >
 >4) How should an LSR handle the case when a label release  bearing a
 >loop detection information arrives.  Taking the same action as is done
 >for "No Route" case doesn't seem right.
 >

can u please point, where it is stated so, I am not able to find it out in
the rfc. Perhaps you are refering to A.1.11., it is not stating actions
pertaining to label release message, so I think the section A.1.4 should
be followed.

I think a note at the end of A.1.11 can be added to point out label
handeling case if at all or the note at the end of A.1.11 can be appended
with following sentence.
If the Loop Detected Status Code comes in Label Release message, follow
the algorithm for the Label Release message instead of No Route
notification. (I don't think this is even necessary as it is obvious)

 >5) When a loop detection notification arrives in response to a request
 >message,  the LSR receiving this message defers this request and sends
 >it repeatedly with the same attributes. Now, if the request attributes
 >cause the nexthop to send a loop detction message (maybe due to path
 >vector or hop count attributes),
 >then what is the use of sending the same message again.
 >

hmm thats interesting. Sorry, I cannot think of any good answer. The only
reason why the attributes are resent is that one does not know the exact
reason for loop detection. So the attributes are sent in the hope that
this time they will be correct. Implementation can take care of such
things where it makes a check at outgoing Path Vector while sending the
message, and stop it from delievering if it indeed finds the address of
the LSR to which it is being sent.( Thanks Eric for similar explaination
in some previous mail. )


 >thanks in advance
 >manoj

-abhijit

-- 
Abhijit Gadgil
Graduate Student,
Dept. of Electrical Engineering,
IIT, Bombay.