Re: [Idr] draft-bashandy-idr-bgp-repair-label-01: when does PE revert back to standard label?

[Jakob Heitz <[e-mail]>]

Robert,

You are talking about something quite different.
Please understand the scenario I described
and answer it directly.

I'll describe it again, using the diagram from the draft:
I will add another CE, CE2. CE2 will send an IP packet to CE.

Initial topology:

         +--------------------------+
         |                          |   +-----CE2
         |      BGP free Core       |  /
         |                          | /
         |      +------------------PE1----+
         |     /                    |      \
         |    /                     |       \
         |   /                      |        \
         |  /                       |         \
         | /                        |          *
        PE3                         |          CE....... VPN prefix
         | \                        |          *          (P/p)
         |  \                       |         /
         |   \                      |        /
         |    \                     |       /
         |     \                    |      /
         |      +------------------PE2----+
         |                          |
         |                          |
         +--------------------------+

CE detaches from PE1 and attaches to PE3.

         +--------------------------+
         |                          |   +-----CE2
         |      BGP free Core       |  /
         |                          | /
         |      +------------------PE1
         |     /                    |
         |    /                     |
         |   /                      |
         |  /                       |
         | /                        |
        PE3                         |          CE....... VPN prefix
       / | \                        |          *          (P/p)
      +  |  \                       |         /|
      |  |   \                      |        / |
      |  |    \                     |       /  |
      |  |     \                    |      /   |
      |  |      +------------------PE2----+    |
      |  |                          |          |
      |  |                          |          |
      |  +--------------------------+          |
      |                                        |
      +----------------------------------------+

If PE3 sends an MPLS packet to PE1, PE1 uses the repair label to send to PE2.
If CE2 sends an IP packet to PE1, PE1 should use the primary label to send it to PE2.
It should NEVER use the repair label to send that packet.

The reason? If the link PE2 - CE were to fail, PE2 would not use the backup to PE3.

         +--------------------------+
         |                          |   +-----CE2
         |      BGP free Core       |  /
         |                          | /
         |      +------------------PE1
         |     /                    |
         |    /                     |
         |   /                      |
         |  /                       |
         | /                        |
        PE3                         |          CE....... VPN prefix
       / | \                        |          *          (P/p)
      +  |  \                       |          |
      |  |   \                      |          |
      |  |    \                     |          |
      |  |     \                    |          |
      |  |      +------------------PE2         |
      |  |                          |          |
      |  |                          |          |
      |  +--------------------------+          |
      |                                        |
      +----------------------------------------+

--
Jakob Heitz.
 

> -----Original Message-----
> From: Robert Raszuk [mailto:[e-mail]] 
> Sent: Saturday, April 02, 2011 2:31 PM
> To: Jakob Heitz
> Cc: idr
> Subject: Re: [Idr] draft-bashandy-idr-bgp-repair-label-01: 
> when does PE revert back to standard label?
> 
> Jakob,
> 
> >>> PE1 gets no new update from PE2, therefore will continue 
> to use the
> >>> repair label to PE2.
> >>
> >> No update is required to be received from PE2. Loosing 
> PE1's directly
> >> attached path and recalculating all information already present in
> >> control plane is already sufficient.
> >
> > That causes the repair label to be used towards PE2.
> 
> No. You need to realize that control plane trigger is not 
> equal to data 
> plane trigger. Those are two totally separate message paths with 
> different forwarding outcome.
> 
> Any sort of FRR uses data plane/forwarding event to trigger 
> really fast 
> prefix independent switchover (usually less then 50ms) to a 
> backup path.
> 
> However control plane is much slower. But will eventually in 
> 100s of ms 
> converge completely removing the original directly connected 
> best path 
> and it's protection via PE2.
> 
> It is really very simple.
> 
> >>> Ahmed clarified it for me like this: If PE1 receives a packet
> >>> destined for a prefix advertised by CE, PE1 will push the primary
> >>> label advertised by the primary egress PE (be it PE2 or any
> >> other PE)
> >>> and then tunnel the packet to that PE.
> >>
> >> Yes, but only after the local control plane convergence happens.
> >
> > Not necessary.
> 
> ;)
> 
> >> Nope. It does not matter what type of packets are received.
> >> Described in the other (non orthogonal) mail.
> >
> > Not described.
> > Please read carefully and understand before answering.
> 
> I did, but apparently my answer was a little bit implicit but I hoped 
> you will parse it. So let me try again ...
> 
> The draft says black on white:
> 
> "and associate a local label with each prefix such as L3VPN [9], 6PE 
> [10], and Softwire [8]"
> 
> it directly means to me that as written it is not applicable 
> to IPv4 and 
> IPv6 non mpls protection.
> 
> And as suggested that can be fixed extending the scope of the 
> draft to 
> make it applicable also to unicast IPv4 and IPv6 as this is not only 
> safe, but also desired in mpls networks where best external 
> is injected.
> 
> R.
>