From: Howard C. Berkowitz (hcb@gettcomm.com)
Date: Sun Sep 29 2002 - 14:55:47 GMT-3
At 10:32 AM -0400 9/29/02, Dean Whitley wrote:
>At first glance I would say to use the weight for incoming routes, to set
>your preference.
>
>
>-----Original Message-----
>From: nobody@groupstudy.com [mailto:nobody@groupstudy.com]On Behalf Of
>Sam Ye
>Sent: Sunday, September 29, 2002 5:54 AM
>To: ccielab@groupstudy.com
>Subject: how to select a RR as perfer RR to receive route from it, and
>another router as backup
>
>
>Hi all:
> In a redundant RR enviroment, how can I select a RR as primary router
>and another as backup. thanks
Again, this depends if it's test practice or real world. In the real
world, I think preferring an RR in a cluster is a bad idea.
Now, let's talk about some test characteristics, and then return to
the real world. Let's try some ugly ASCII art:
Core1.........Core 2 (Core 1 and Core 2 have iBGP to RR1/RR2
v v as well as to each other)
v v
RR1========RR2
| | | |
| +------+ | |
| | | |
RC1 RC2 |
| |
+------------+
Assume RR1 is primary and RR2 is secondary. If a route comes in to RR2, you
can't use weight, because that wouldn't propagate to the clients.
Setting LOCAL_PREFERENCE might work in routes sent to the clients.
The clients, in turn, would set different LOCAL_PREF to the RR's for
routes they receive.
But what if RC1 and RC2 have eBGP peers? Well, you could again use
local pref. This could, however, get ugly if you have other clusters
or eBGP speakers talking to the same external AS, and the local_pref
values aren't coordinated.
It would also get ugly if you were using hierarchical route reflectors.
So, yes, you probably could do it, but it would be very hard to scale
in a complex network. It doesn't defeat the purpose of a cluster,
but goes against its spirit.
This archive was generated by hypermail 2.1.4 : Mon Oct 07 2002 - 07:44:07 GMT-3