[c-nsp] ebgp load balancing using maxiumu-paths TCAM impact on Sup720-3BXL?

[Gert Doering]

Hi,

On Thu, May 21, 2009 at 10:54:40AM -0700, Peter Kranz wrote:
> I have two edge routers, with each edge router has a BGP session to the same
> upstream provider (Level3 AS3356)
> The edge routers are connected 
> I would like outbound traffic from our AS that arrives at either edge router
> to be load balanced across the two sessions to AS3356
> 
> I tried adding bgp multipath and ibgp multipath to install both routes and
> load balance, but this is not working as of yet.. and perhaps I am going
> about this the wrong way..

Following up on this.  I don't think this is going to work the way you
envision it - one of the practical problems is that if "Router A" sends
a packet to "Router B" (to be balanced to the other upstream link), you
might up with "B" sending it *back* to "Router A", because it has 
hashed the parallel routes differently, or because it does per-packet
balancing and will always send 50% of the packets back to "A".

So the chance of routing loops is pretty high.


Regarding actual implementation: as "eBGP" will always win over "iBGP",
there isn't anything you can do to make this "multipath" - this would
only work if both paths are "eBGP" or both "iBGP".


To solve your issue at hand - what we do in this situation is to 
play with MED a bit.  We have a similar setup, two routers on our
end and two provider routers on their end, with same-sized links
on router pair "A" and "B".  In our case, what we did was to move
all traffic to "<upstream>_3320_" (the local incumbent) to router "A",
and to move everything else to router "B".

We do this by manipulating the MEDs

  Router A:  MED for "all paths that do not have 3320 in them" +50
  Router B:  MED for "all paths that *do* have 3320 in them" +50

for us, this means that the traffic is somewhat balanced, and has the
added benefit of telling use exactly how much traffic goes to 3320...
(which is always a big problem in germany).

If A or B fails, the MED manipulation is not strong enough to force 
traffic away from the other link (as would lowering the local-pref be),
so we still have full redundancy.


So, in your environment, you could find some other characteristic that
would balance your traffic somewhat, like "peers of my upstream" vs.
"upstreams of my upstream" (which is usually tagged by BGP communities)
and manually balance the traffic that way.  Try and error...

gert

-- 
USENET is *not* the non-clickable part of WWW!
                                                           //www.muc.de/~gert/
Gert Doering - Munich, Germany                             gert at greenie.muc.de
fax: +49-89-35655025                        gert at net.informatik.tu-muenchen.de