Re: [Idr] draft-walton-bgp-add-paths-06.txt as IDR WG document

> -----Original Message-----
> From: Danny McPherson [mailto:danny@tcb.net] 
> Sent: Tuesday, November 25, 2008 3:09 PM
> To: PAPADIMITRIOU Dimitri
> Cc: Yakov Rekhter; idr@ietf.org
> Subject: Re: [Idr] draft-walton-bgp-add-paths-06.txt as IDR 
> WG document
> 
> 
> On Nov 25, 2008, at 6:13 AM, PAPADIMITRIOU Dimitri wrote:
> >
> > if an id is prepended to the best path, and only the latter is
> > advertized this would not incur more instability than today's  
> > situation.
> > except if for whatever reason the BGP speaker decides to 
> modify that  
> > id
> > (if my reading is correct nothing prevent this from the current
> > write-up)
> >
> > when multiple paths (each prepended with an id) are progressively
> > advertized you have an effect on and then each would be 
> progressively
> > removed (A:ID1, A:ID2,..., A:IDn) and then (W:ID1, W:ID2,...,W:IDn)
> > compared to the current situation where after the initial  
> > advertisement,
> > a sequence of increasing length path is advertized followed by a
> > withdrawal or stabilization to a newly preferred path. So 
> the effect  
> > is
> > the delta on the withdrawn sequence from W:ID1,...,W:IDi 
> (in the work
> > case, i=n i.e. the full sequence is withdrawn) and the delta on the
> > advertized sequence A:IDj,...,A:IDn (in the worst case j=2 i.e. one
> > would stabilize on the second path)
> 
> I mean between adjacent nodes, not system wide.  

it is this behaviour b/w adj speakers that i tried to explicitly 
characterize.

> More paths means more state, more paths means more churn.  This 
> is the obvious trade-off here when employing add-path.

you can then translate this into state maintenance/processing 
overhead.

-d.

> -danny
> 
> 
_______________________________________________
Idr mailing list
Idr@ietf.org
https://www.ietf.org/mailman/listinfo/idr