From: Nick Shah (nshah@connect.com.au)
Date: Wed Oct 16 2002 - 22:21:09 GMT-3
John
When configuring IBGP peers, the general idea is that they would be
connected with redundancy in mind, so there would be multiple / alternate
paths between each peers. Now if you configure them with just their directly
connected interface ip address as peers, if that interface goes down, bang
the peering goes down.
So what do you do ? You run IGP's between them so that all their ip
addresses are known (which should be straight forward) and then you run
peerings between loopbacks. Since the loopback as an interface never goes
down (unless the whole box goes down) as long as there is atleast one path
active between the peers, the session remains up.
All fine and good with IBGP peers. You dont need to mention multihop, but
you need to mention "update source loopback0"
However if similar stuff is done with EBGP peers, say they have redundant
links between them, and you run BGP peering between loopbacks then they are
said to be "Not Directly Connected" , obviously because the 2 loopbacks of
respective routers are not directly connected to each other.
So you run have to run a MULTIHOP session between them. Here you need
ebgp-multihop <n>
multihop sessions are also useful in production environments where there
would be some transit router between the Provider and Customer not running
BGP, but routing info is still passed (via statics/IGP), so both the routers
know about their respective remote addresses. Then you run EBGP multihop
sessions between BGP capable routers on both ends. This is not a very common
practise, but I have seen /done them in a few cases.
rgds
Nick Shah
----- Original Message -----
From: "Paglia, John (USPC.PCT.Hopewell)" <JPaglia@NA2.US.ML.com>
To: <rem@digdomsol.com>; <ccielab@groupstudy.com>
Sent: Thursday, October 17, 2002 9:43 AM
Subject: RE: EBGP Multihop's necessity with loopback addresses
> Thanks for the reply. Actually, I am not having a problem. I was just
> wondering about something someone told me. I never have configured
'ebgp-mu'
> for connected neighbors, but this person said 'you need it if using
> 'update-source loop 0' because loop 0 isn't directly connected to the
> neighbor.'
>
> I just wanted to know if anyone else has heard of such madness. I've never
> configured BGP like this and it has never been a problem.
>
> Thanks,
> John
>
> > -----Original Message-----
> > From: Rem [SMTP:rem@digdomsol.com]
> > Sent: Wednesday, October 16, 2002 7:27 PM
> > To: Paglia, John (USPC.PCT.Hopewell)
> > Subject: RE: EBGP Multihop's necessity with loopback addresses
> >
> > Are you running an igp between those neighbors?
> >
> > Where ebgp-multihop comes into play is between neighbors that are not
> > directly connected and run only ebgp between them, ie your 2 loopback
> > interfaces. if you run a trace between these 2 while running only ebgp
you
> > will see that they will not be able to perform a tcp connection, that is
> > because the loopbacks do not know how to get to the remote side. if an
igp
> > is running between your serial connection then the route is established
> > and
> > the tcp connection can occur.
> >
> > try removing any internal routing protocol from your tables and see if
it
> > drops the connection. it has been my experience that this is currently
the
> > case and i know from experience that it has been a problem. watch out
for
> > it
> > on the test when your start redistributing and the igp route goes away
and
> > all of a sudden your bgp drop also. its a nasty little thing if you
don't
> > see it coming.
> >
> > hth
> > Ross
> >
> > -----Original Message-----
> > From: nobody@groupstudy.com [mailto:nobody@groupstudy.com]On Behalf Of
> > Paglia, John (USPC.PCT.Hopewell)
> > Sent: Wednesday, October 16, 2002 1:01 PM
> > To: 'ccielab@groupstudy.com'
> > Subject: EBGP Multihop's necessity with loopback addresses
> >
> >
> > I recently heard that if you are establishing your BGP neighbors using
> > 'update source loopback 0', you should also use the 'ebgp-mu' cmd, even
if
> > the neighbors are directly connected...the reason being that your
loopback
> > is NOT directly connected to the neighbor. However, in my experiments I
> > have
> > never done this for neighbors that are directly connected, yet have
> > established peerings successfully.
> >
> > Is there validity to this statement, and if so, under which
circumstances
> > is
> > it absolutely vital, other than the 'non-physically or nbma topology'
> > scenarios??? Something tells me that this may be an older IOS issue or
> > something like that.
> >
> > John
This archive was generated by hypermail 2.1.4 : Tue Nov 05 2002 - 08:35:48 GMT-3