From: Jeremy Gray (jeremy.gray@xxxxxxxxxx)
Date: Sun Feb 04 2001 - 08:38:29 GMT-3
Here's a few tips.
Next-hop-self - This typically being for IBGP sessions. EBGP will have your
outbound interface as the update-source for each update by default.
You simply set this instead of advertising the subnet of the EBGP peer(s)
into your IGP.
For example lets say you're an ISP at LINX. You can either advertise the
linx subnet in your IGP (redist connected, etc ) ; so that your IBGP peers
who receive this update have the source (the external LINX LAN) subnet in
their routing table and can therefore treat the EBGP routes from any LINX
peers as valid (since the next hop reachable)....
OR...
You can set next-hop-self on the IBGP sessions to your IBGP peers on the
assumption they know how to get to you... Because your source address is
known or adverised in the IGP already.
When you may not to use NEXT-HOP-SELF on an IBGP session...
If you have a pair of routers on a shared NAP (say MEA-EAST1 and MEA-EAST2)
it may be more efficient to retain the EBGP next-hop when you advertise
these routes to your IBGP peers.
This is becuase ...When packets are then routed towards these routes they
can then flow back to both of you border routers rather than 'one'. For
example even if the original EBGP peering session was on your MEA-EAST1
router, they can leave via MEA-EAST2, if the external nexthop route being
followed via each routers recursive lookup used the best metric back to
'mae-east subnet' instead of a specific host address. This is of course
touching on design issues...
When you don't care about the next-hop...
If the EBGP peer is via a point-to-point subnet it makes no odds.. but you
still MUST do one or the other, route the EBGP peers subnet or use
next-hop-self.
SYNC.
A somehwat inappropraite default these days... Use this only if your EBGP
peers are seperated with none-BGP router hops in your AS. And then
redistributed the BGP routes into the IGP, the tow statements really go
together SYNC = (redistributed into you IGP). Clearly this has limited
value in the real world. The internet now has over 90k routes and even the
most robust IGP's are going to have trouble. In reality if you had to do
this with teh entire internet, you'd need big-monster routers anyway - so
you may as well run BGP on them, then you're back to "NO SYNC", since you
really don't want to kill your IGP.
In a lab scenario ... or other non INTERNET-SCALE network...
EBGP routers are seperated by none-BGP routers. The none-BGP routers need to
have the same routes, so that if a BGP router forwards a packet via one of
these routers the receiving router has somewhere they can send it (other
than the bit bucket). The problem is that the EBGP router needs to know
when the none BGP routers "are ready" to forward packets. Therefore they
wait until they can see the routes in the IGP before they do. And thus they
are then "synchronised" with teh IGP and can actually use the BGP routes to
forward packets. Likewise they can then also advertise all synchronised
routes to other EBGP peers as each route becomes 'synchronised'.
NO SYNC.
All routers between EBGP routers run BGP. THerefore packets routed across
your network will be routable on each hop, since each hop as the full BGP
routing table. No need to check if the route is in the IGP, coz its not.
The the real world this is what you'll use.
Update-source.
Its a good common practice to set update-soure between IBGP peers to the
loopbacks and this is useful when your routers have multiple interfaces
because it creates resilience if an interface fails. The neighbor statement
to the to remote peer therefore points to your loopback address, and your
neighbor statement points to theirs and you both use update-source loopbackN
Hopefully this clarifies rather than complicates....
Regards, Jeremy.
-----Original Message-----
From: nobody@groupstudy.com [mailto:nobody@groupstudy.com]On Behalf Of
Fred Ingham
Sent: 04 February 2001 03:06
To: Alan Basinger
Cc: CCIE Newsgroup (E-mail)
Subject: Re: BGP Next hop self
Alan: On the r1 neighbor statement only, assuming that r1 does not have
the subnet between
r2 and r3 in its routing table. You could also redistribute the r2-r3
subnet into the r1-r2
IGP. What did you see when you tried this out?
Fred.
Alan Basinger wrote:
>
> Hi all,
> With only 2 weeks till me second attempt I am now in the panic mode.
> Here is my question.
>
> r1 ------- r2 ------- r3
>
> r1 being in bgp as 10 and r3 being in bgp as 20 and r2 is the border
router.
> Would I use the bgp next hop self command on r2 pointing to both r1 and
r3?
>
> Alan Basinger
> Systems Engineer
> SBC DataComm
> Houston Texas
> abasinge@swbell.net
>
> | |
> ||| |||
> .|||||. .|||||.
> .:|||||||||:.:|||||||||:.
> C i s c o S y s t e m s
> Certified Gold Partner
>
This archive was generated by hypermail 2.1.4 : Thu Jun 13 2002 - 10:28:35 GMT-3