From: James (james@towardex.com)
Date: Sun Aug 08 2004 - 15:01:37 GMT-3
On Sun, Aug 08, 2004 at 12:20:05PM -0400, Brian McGahan wrote:
> Another reason the BGP update source can be changed is to offer
> a very basic form of load balancing. For example, if you have multiple
> circuits from the same provider terminating on the same router, you can
> use BGP's dependence on the IGP table to perform load balancing. Take
> the following example:
>
> /---\
> R1 R2--ISP
> \---/
>
> R1 has multiple T1 circuits to the ISP which terminate on R2.
> Suppose the interfaces are 12.0.0.0/8 and 21.0.0.0/8. R1 and R2 also
> have the loopback addresses 1.1.1.1/32 and 2.2.2.2/32 respectively. A
> BGP peering is created using these loopback addresses, and identical
> routing information is configured for the loopback addresses out both
> interfaces:
Brian,
Thanks for the great example here! :)
One thing I might want to add to your example is as follows.
One case of issue that I've had in the past (I am sure others can chime in too)
is when you use certain medium that does not always detect that circuit is
down, the glued down static route to the next-hop will still prefer the downed
path.
We had this case at my previous employer's network. We had 3 ATM DS3 transit
connectiosns to our ISP, using load balancing with ebgp-multihop 2 setting
3 equal-cost static routes to the next-hop. It worked most of the time, even
when one of the circuits go down. What started happening after that is sometimes
our provider's ATM cloud starts having problems deestroying the PVC, but for
strange reason, the PVC never shows down to our end, neither provider's edge
router end, causing routers on both ends to think the circuit is still up and
passing number of packets to a downed link.
In case of solving that problem, we've had two major choices:
o Use ATM OAM management down the PVC so that it will shut it down when PVC
becomes unhearable.
o Get rid of multihop BGP, setup 3 seperate singlehop sessions and use
'maximum-paths 3' under router bgp process. By using single hop session
per sub-interface, if PVC stops responding, the respective session's
hold down timer will eventually expire and tear down that session.
We ended up using OAM on the pvc's mainly b/c of two reasons. We didn't want
to put more CPU burden on our router. When one of the circuits go down with
single hop sessions, one of those sessions will tear down, making BGP take more
CPU during the process. And secondly, our upstream didn't really want to
enable 'maximum-paths 3' globally under their router bgp process since they
were afraid of potential breakage for some other customers, which indeed is
completely understandable. OAM happened to work very well for us. Whenever a
PVC started acting up weird again, it always would shut it down, saving us from
further troubles and annoyance w/ packet loss! :)
Just my experience with OAM vs. BGP multipathing.
-J
-- James Jun TowardEX Technologies, Inc. Technical Lead Network Design, Consulting, IT Outsourcing james@towardex.com Boston-based Colocation & Bandwidth Services cell: 1(978)-394-2867 web: http://www.towardex.com , noc: www.twdx.net
This archive was generated by hypermail 2.1.4 : Fri Sep 03 2004 - 07:02:35 GMT-3