[email protected]
[Top] [All Lists]

Re: Adopting draft-francois-ordered-fib-00.txt as a WG document

Subject: Re: Adopting draft-francois-ordered-fib-00.txt as a WG document
From: "Joel M. Halpern"
Date: Mon, 21 Nov 2005 17:19:19 -0500
Remember that "protected by a fast reroute protection technique" may be on a prefix by prefix basis (protected for some, but not for others) and may be hard to determine.
If this were restricted to
o link up
o metric change
then I would not be as uncomfortable. Those are indeed generally cases where care is more important than speed.
Joel M. Halpern

PS: Orderly shutdown would then be a manual metric change followed later (a few seconds later would suffice) by abrupt removal.
At 05:01 PM 11/21/2005, Olivier Bonaventure wrote:

> I am really uncomfortable with changing the general convergence behavior
> of the routing protocols to handle the case when the human knows he will
> shut things down.  If this were only triggered by a special
> advertisement that indicated orderly shutdown, that might be less
> bothersome.

Of course, the ordered FIB update would only be used for non-urgent
topology changes. Those topology changes include :
- link up event (all link up events are non-urgent)
- metric change event (going up or going down - all metric changes are
- manual shutdown of a link (could be explicitely indicated by first
setting the metric to almost infinity and later removing the link)
- failure of a link which is protected by a fast reroute protection
technique (again, if the link is protected, then its metric can be set
to a large value)

> Particularly, it seems strange to be talking about slowing convergence
> when we are repeatedly being asked to speed up convergence.

Note that with the completion messages described in the draft, the
convergence time can be reasonable. Our simulations indicate that
sub-second ordered convergence is possible with completion messages.
These simulations are not yet published, but if there is some interest
we could prepare a short technical with our findings.


Rtgwg mailing list
[email protected]

<Prev in Thread] Current Thread [Next in Thread>