The MPLS WG Archive

["Adrian Farrel" <afarrel@movaz.com>]

Hi,

I have some comments which are mainly requests for clarification to be added to the draft.  I'm sorry to raise them at this late stage, but better now than even later...

 

1. It would be nice if you made it clear up front that

   this draft applies only to DU.  Currently this is

   tucked away in para 3 of section 6.

 

2. In the first line of 6.1.1 could you say that "the

   Mapping message" is a "newly received Mapping message"?

 

3. It would be helpful to add some text to cover processing

   of other messages while an LSR is in the process of

   restarting: in particular Withdraw and Release.  Although

   this is pretty obvious, I believe you run the risk of

   failure to interop correctly unless you spell it out: viz.

   Withdraw should match in table, delete from table and

   send Withdraw upstream as required

   Release should match in table, if entry is stale should

   ignore, if entry is not stale should process according

   to normal Release procedures on local node.

 

4. It would be nice if you exposed that all Address messages

   need to be exchanged before the downstream node starts

   resending Mapping messages.  Section 6.1.1 is based on the

   assumption that these messages have been received.

 

5. Is it normal procedure to give suggested values for all

   timers in drafts?

 

6. Section 6 states that there is an assumption that IP

   forwarding state is preserved in parallel to MPLS

   forwarding state.  I looked hard for a reference to this

   in the text and only found the case of the restarting

   Egress LSR needing to look up the next hop of a FEC if

   it is also configured to generate a non-null, unique

   label for such a FEC.  Further this only applies if the

   FEC is of the type that has its next hop determined

   through the IP forwarding table.

   This seems much weaker than the statement in section 6.

 

As a general question, has anyone done any analysis of how long it would take to redistribute all of the addresses and labels on a router in a large network?  In other words, what is a reasonable bound for the Recovery Time?  (Note that the neighbors can't start re-using labels until the Recovery Time is complete.)

 

Hope this is of some help.

Regards,

Adrian