The MPLS WG Archive

["David Allan" <dallan@nortelnetworks.com>]

Title: RE: [Fwd: I-D ACTION:draft-pan-rsvp-te-restart-01.txt]

Ping:

I don't see control and data planes as tightly coupled in your proposal. You're clearly operationally divorcing the two (which is the trend you refer to for BGP, IGP and now RSVP restart mechanisms). Not that this is a bad thing, it just cannot be done in isolation. You (and others, my point is not specific to this I-D) are eliminating all forms of fate sharing between the control and data planes, despite (in the MPLS case vs. GMPLS) the use of common resources in the control and data plane adjacencies.

This may suggest that control plane failure is today significantly more frequent than data plane or link failure and there is a real problem to solve, but adding restart mechanisms to everything in the control plane is not the complete answer. Every divorce in the fate-sharing world adds a new failure detection requirement and more maintenance messaging.

cheers
Dave

-----Original Message-----
From: Ping Pan [mailto:pingpan@juniper.net]
Sent: Wednesday, August 15, 2001 10:33 AM
To: Allan, David [CAR:NS00:EXCH]
Cc: neil.2.harrison@bt.com; mpls@UU.NET
Subject: Re: [Fwd: I-D ACTION:draft-pan-rsvp-te-restart-01.txt]

> David Allan wrote:
>
> Ping:
>
> I'm having trouble correlating route flap instability data with your
> problem statement. Similarly I have trouble with the analogy between
> graceful restart for routing (which has a fair chasm between it and
> forwarding) and a soft state signalling protocol which would appear to
> require a much tighter coupling.
>

Dave,

The point was that in the network, there could be cases where the
control engine is crashed while the data forwarding engine is fine. To
reduce the side-effects, control protocols (routing and signaling)
should use various graceful restart mechanisms. In our draft, I think
the control and forwarding are tightly coupled. Please let me know why
you think otherwise.

> Is this really more applicable to GMPLS and should be tagged as such?
>

Both GMPLS and MPLS can benefit from this. In case of GMPLS, graceful
restart can be used to recover from control channel failure.

Regards,

- Ping