The MPLS WG Archive

[Eric Gray <ewgray2k@netscape.net>]

Title:

Leonardo,

    Thanks for being persistent.  :-)

    See below...

--
Eric

Leonardo Balliache wrote:
> Eric,
>   
>   
> ...
> SNIP ...
>   
>   
> >    The reason why the procedure is not meant
> > to be pseudo code is that
> >     
> > it describes what it is that the implementation is supposed to do.
> > There is
> >     
> > no strict ordering or contextual mapping of the things your
> > implementation
> >     
> > has to do to the way in which your implementation does them.  If you
> > take
> >     
> > a broad look at how many ways there are to do the rest of the things
> > that
> >     
> > any label switching implementation might do, you will quickly realize
> > how
> >     
> > pointless it would be to try to cover all the ways in which this
> > particular
> >     
> > set of "steps" might be done.
> >     
> >   
>   
> I agree, but perhaps the specification shouldn't be so specific.
> Working and revising 
> the spec and coding it I have realized that you are right. Many
> questions are answered 
> when one understand better the LDP logic.
>   
>   
> Talking about "Recognize New FECs", I insist that:
>   
>   
> 1.- In "unsolicited independent control" function
> "prepare_label_mapping_attribute" is 
> called. This function requires the argument "propagating_" which is set
> (before calling 
> the function) to true if the LSR has received previously a label
> mapping for fec from 
> nexthop. It is set to false otherwise.
>   
>   
> 2.- In "unsolicited ordered control" the function
> "prepare_label_mapping" is also called. 
> But this time the argument it needs (propagating_) is not previously
> set. Can I suppose
> that it should be false if the LSR is the egress for that fec and true
> if the LSR has received 
> previously a label mapping for fec from nexthop?
>   
Okay, good question. In the Label Request case (LRq.9), careful reading
will show that the
value passed to a prepare label mapping attributes function is the
"constant" isPropagating
in the Ordered Control mode (sub-step 2). This step is not reached if 
the local LSR is 
not either the egress, or in receipt of a previous label from its next
hop for the FEC.

The wording difference is subtle, and easy to miss - but once you catch
the difference,
it is clear.

The same applies in processing a Label Mapping (LMp.20, LMp.24 and
LMp.29). You 
would only reach one of these steps if A) you are currently in the
process of processing a
Label Mapping (in which case you are - by definition - propagating).

However, in the "Recognize New FEC" case - specifically FEC.1,
Downstream Unsolicited
Ordered Control, substep 2 - the value of the "variable" Propagating is
not set (as it was in
Downstream Unsolicited Independent Control).  This is a good catch on
your part, and may
be an indication that either nobody has run into this mode (as is true
for me), or they were 
able to figure it out (as is the case for you).

In fact, the setting of "Propagating" should not be part of the
iteration, since the local LSR
has only one next hop for the given FEC, no matter how many upstream
peers it may have.
That being the case, it is silly to set this "variable" at each
iteration.  Also, as a result of that 
consideration, it is clear that all of the steps in FEC.1, Downstream
Unsolicited Ordered
Control, should be skipped if the local LSR is neither the egress nor
in receipt of a Label
Mapping from its next hop.

Ina should fix this either
by removing part of this section, or by changing it to read - 

      FEC.1   Perform LSR Label Distribution
procedure:

            For Downstream Unsolicited Independent Control

               1. Has LSR previously received and retained a label
                  mapping for FEC from Next Hop?
                  If so, set Propagating to IsPropagating.
                  If not, set Propagating to NotPropagating.

               2. Iterate through 5 for each Peer.

               3. Execute procedure Prepare_Label_Mapping_Attributes
                  (Peer, FEC, InitAttributes, SAttributes, Propagating,
                  Unknown hop count(0)).

               4. Execute procedure Send_Label (Peer, FEC, SAttributes)

               5. End iteration from 1.
                  Goto FEC.2.

            For Downstream Unsolicited Ordered Control

               1. Has LSR previously received and retained a label
                  mapping for FEC from Next Hop?
                  If so, set Propagating to IsPropagating and Goto 3.
                  If not, set Propagating to NotPropagating.

               2. Is LSR egress for the FEC?
                  If not, Goto 6.

               3. Iterate through 5 for each Peer.

               4. Execute procedure Prepare_Label_Mapping_Attributes
                  (Peer, FEC, InitAttributes, SAttributes, Propagating,
                  StoredHopCount).

               5. Execute procedure Send_Label (Peer, FEC, SAttributes)

               6. End iteration from 1.
                  Goto FEC.2.

            For Downstream On Demand Independent Control OR
            For Downstream On Demand Ordered Control

               1. Goto FEC.2.  (See Note 2.)
> ... 
> SNIP ...
>   
> > --
> >     
> > Eric
> >     
> >   
>   
> Best regards,
>   
>   
> Leonardo Balliache
>   
>   
>   
>   
>   
>   
>   


_______________________________________________
mpls mailing list
mpls@lists.ietf.org
https://www1.ietf.org/mailman/listinfo/mpls