Re: [nsp] BGP load balancing depends of bandwitdth

From: Philip Smith (pfs@cisco.com)
Date: Mon Jun 24 2002 - 19:38:16 EDT


Hi,

At 15:09 24/06/2002 -0400, barney - SPAMs wrote:
>Our company have connectivity from two independent ISPs, ISP A and ISP B.
>Both links are 10 Mbps fullduplex
>
>Border router is Catalyst 6509 with hybrid IOS Version 12.1(11b)E.
>
>My problem is that almost whole traffic is going across ISP B, and very
>small traffic is going across ISP A.

When you say that almost all your traffic goes across ISP B, which
direction? I suspect both directions, otherwise you might have mentioned
this imbalance?

There isn't anything "wrong" with your configuration as such, but you will
have to do a little bit more work to try and balance the traffic out.

It is quite possible that your neighbour ISP A is further from the "centre"
of the Internet than ISP B, so inbound traffic sees a better path to you
through ISP B. One way to help with this is to announce your networks via
ISP B with a longer AS-PATH than you do via ISP A. Rather than doing an
AS-PATH prepend at random, go to a looking glass (nitrous.digex.net is a
good place to start) and see how your nets are announced. If you only see
your nets from ISP B, then try a single AS prepend to ISP B, and see how
that goes. And repeat until you either see both paths at the looking glass,
or you see ISP A path only... Be careful with resetting your BGP session
several times though - your prefixes will be damped by upstream ISPs. I'd
imagine that you'd only need one or two AS prepend anyhow - very few ISPs
need any more than that. Your configuration for the AS-prepend I've
annotated below.

As for outbound traffic, my guess is this might be less of an issue for
you. But in case it is, this depends on what prefixes you get from your
upstream. I can't guess from your configuration - you may be receiving a
full routing table from both (not a recommended practice - you don't need
to know the full detail of the Internet just to loadshare between two
upstreams). There are many possible solutions here, best look at the
tutorial I gave at NANOG for some examples:
www.cisco.com/public/cons/seminars/NANOG25/BGP-Tutorial.pdf

>My configuration:
>
>router bgp 11111
> no synchronization
> no bgp fast-external-fallover
> bgp log-neighbor-changes
> network 1.1.0.0 mask 255.255.224.0
> network 2.2.0.0 mask 255.255.224.0
> neighbor 7.7.7.7 remote-as 2222
> neighbor 7.7.7.7 filter-list 3 out

neighbor 7.7.7.7 route-map ISP-A-out out

> neighbor 8.8.8.8 remote-as 3333
> neighbor 8.8.8.8 ebgp-multihop 4
> neighbor 8.8.8.8 update-source Loopback0
> neighbor 8.8.8.8 filter-list 3 out

route-map ISP-A-out permit 10
  set as-path prepend 11111 11111

>
>1.1.0.0 and 2.2.0.0 - our networks
>7.7.7.7 - neighbor of ISP A
>8.8.8.8 - neighbor of ISP B
>
>I want to configure load-balancing between ISP A and ISP B depends to
>bandwitdth. Could someone send example of configuration ?
>
>PS: sorry for bad english :-)

There isn't any function available to configure it depending on the link
bandwidth. You will have to do this manually. I find most organisations who
are fine tuning loadsharing are monitoring their link performance on a
daily basis, and probably making any necessary adjustments to loadsharing
on a weekly basis. It isn't too hard to do - and is generally part of the
function of a network operations team anyhow.

Hope this helps!

philip

--



This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:01 EDT