From: Fred Nielsen (fred_nielsen@xxxxxxxxxxx)
Date: Sat Feb 17 2001 - 21:03:48 GMT-3
Chuck, the ASCII art didn't look good here, but oh well. Here is the best
way I can think to describe the "feature":
You have one AS consisting of routers 1-5:
1 2
3
4 5
Routers 1 and 2 are each separately connected to the same external AS, and
both learn the same route from that AS. Both 1 & 2 have an IBGP session
established with 3 and both send the route from the external AS to 3. 3
sees both in it's BGP table and selects only one as best, as expected.
Here was the interesting part. 3 is a route reflector to 1, 2, 4 & 5. Only
the route that 3 selected as best was being seen on 4 & 5, not both. So 4 &
5 only showed one instance of the external route. Hmmm.. If we had been
running a full mesh, 4 & 5 would have seen both routes directly from 1 & 2.
But we were using reflection, the assumption being that reflection would
accomplish the same end state. It was not.
Now, I did some more looking at this and remembered an interesting section
in Halabi's book regarding route reflectors and peer groups. It just so
happens that all the neighbors on router 3 were configured in the same peer
group. In Chapter 8, Controlling Large Scale Autonomous Systems, there is
section entitled Route Reflectors and Peer Groups. Essentially all the
route reflector clients within a peer group need to be fully meshed among
themselves, according to the book. The explanation of the potential problem
is somewhat tricky to wrap your mind around, but this may have something to
do with it.
The other thing I'm wondering, I know only the routes selected as best are
sent to the Output Policy Engine by BGP (for both internal and external
peers), but does this behavior also apply to reflection of updates between
RR clients?
I'm going to tweak some more and see what I can figure out next week.. Time
to go play in the mud.
-Fred Nielsen
-SBC Datacomm Design Engineer
-CCIE No. 6276
-[fred_nielsen@hotmail.com]
----Original Message Follows----
From: "Chuck Larrieu" <chuck@cl.cncdsl.com>
Reply-To: "Chuck Larrieu" <chuck@cl.cncdsl.com>
To: "CCIE_Lab Groupstudy List" <ccielab@groupstudy.com>
Subject: BGP Route reflection - route choices
Date: Sat, 17 Feb 2001 13:33:15 -0800
I haven't had a chance to check this out myself. But a couple of guys in the
office were working on a remote rack, and posed this apparent discovery.
Hope the ASCII art survives and hope I am remembering this correctly.
Fred - you still monitor this list? If so, correct me if I am wrong) :
|-------Router 2---|----other AS
0ther AS----Router_1 | |
|-------Router 3---|-
| |
|-------Router 4----|
this AS has two paths into the same remote AS. BGP learns the router to
Other AS through both router 1 and router 2.
The route that is reflected to all routers is that of router 1's path to
Other AS, even though in some cases, router two is the better exit point
via the IGP routes )
We could not come up with a rationale for this. Artifact? Other issues with
igp distribution into BGP? Maybe some filters?
Is what they were seeing the result of some design spec within BGP or
Cisco's implementation of route reflection?
Anyone else seen this?
Chuck
A long shot at passing is better than no shot.
Right now that's all I got to get me through,
So I gotta believe!
( paraphrased from Kathy Baille / Baille and the Boys
a song from several years ago )
This archive was generated by hypermail 2.1.4 : Thu Jun 13 2002 - 10:28:52 GMT-3