Re: [Idr] New Version Notification for draft-liang-idr-bgp-flowspec-time-00.txt

[Jeffrey Haas <[e-mail]>]

On Mon, Oct 19, 2015 at 09:50:10PM +0200, Gunter Van De Velde wrote:
> If the purpose is to communicate timing properties attached a NLRI, then why limit it to ‘just’ flow spec AF? it could potentially be used for any type NLRI.
> 
> To understand the potential, could you share the list of those proposals that were not rolled out due to a micro-second BGP signalled timing requirements?
> 
> I am trying to understand if for those BGP signalling would make sense or if there would be better options.

At the moment, I'm generally opposed to this time-based routes for a few
reasons:

1. They're "time bombs", network behavior becomes tricky to troubleshoot
based on some box's perception of time.  If the impact is intended to apply
to multiple routers, having skew of time may do bad things.  Robert
suggested route loops as an example.

2. Incremental roll-out will be painful.  Some routers will apply the route
immediately, others will wait on the time-bomb.

3. Other service providers probably don't want time bomb semantics of
someone else's route.  As it is, the Internet penalizes route instability.
this makes the feature problematic from a transitivity standpoint.

4. In the absence of add-path routes, these routes seem a bit weird anyway.
You're injecting a route that must be selected to become propagated, but
then you have to distinguish selected vs. installable.

-- Jeff