Pedro,

From: "Pedro R Marques" <roque@...>

> On Thu, 2003-07-17 at 12:35, Adam Debus wrote:
> 
> > in theory a RR shouldn't ever see real
> > traffic...just BGP exchanges, right?
> > 
> 
> The theory is that your RRs should be in the traffic path, although it
> is possible to implement them otherwise if you are really careful and
> understand what an RR does.

I am not sure if the RRs are always supposed to be in the traffic path. 

---[A]\-----------
       \         |
       [RR]---- [C]
       /         |
      /          |
---[B]-----------

A, B and C are 3 clients of the RR. RR here reflects the routes from A and B to C - but doesnt
come anywhere in the data forwarding path. C can throug some IGP cloud can send data directly to A
and B. 

Isnt this possible?

And if its isnt, then i would love to know why? 
 
> The issue is that an RR performs path selection according to its
> perspective of the IGP distances.
> 
> So if you where to have for instance a topology such that you have
> 
>      5     10      5   (igp metrics)
>          +------+------+------+
> gw1    gw2     gw3    RR
> 
> Assuming that both gw1 & gw3 have the same external route and that all
> the rest is the same, the RR will pick and reflect the path from gw3,
> which is probably the wrong exit point from gw2 point of view.

This is then a limitation of the RR. Shouldn't we be addressing it? Lately there have been some
proposals in the IETF to advertise multiple BGP routes. Will that not help here?

> 
> Of course the example above is quite artificial, but it should
> illustrate the point.
> 
> Another consideration is that when you have a network that is designed
> as core/area topology, the RR is typically deployed in the area boundary
> router (core attachment). One of the reasons for this is that the RR
> hides paths between the two domains (core/area) and if load-balancing is
> required traffic has to pass through the RR in order to be split into
> multiple paths.

Ditto above.

> 
> My advice would be to default to have the RRs in the traffic path (in
> real routers), unless you really know what you are doing...
> 
> regards,
>   Pedro.


Rasputin

______________________________________________________________________ 
Post your free ad now! http://personals.yahoo.ca

------------------------ANNOUNCEMENT---------------------------------
---------------------------------------------------------------------
           ISPCON FALL 2003 - Santa Clara Convention Center
     October 20-22 - The Definitive Event Wired and Wireless ISPs
                            www.ispcon.com
----------------------------------------------------------------------
----------------------------------------------------------------------

Replies

[isp-bgp] Re: Zebra as Route Reflector, Pedro R Marques