[c-nsp] Re: [j-nsp] load balancing between multiple BGP links

[Oliver Boehmer (oboehmer)]

Alok <mailto:alokdube at hotpop.com> wrote on Wednesday, September 14, 2005
6:15 PM:

> but i would need 2 sessions?

of course. Why not?

> The algorithm I am trying to point out is generic, anytime my FIB has
> 2 ECMP routes, no matter how I configure them, they should ECMP based
> on actual link util..

You mean "link bandwidth", not "link util", right?
The FIB is stupid, it does what its being told by the RIB. So in theory
you could run EIGRP between the two routers, use EIGRP's unequal-cost
load-sharing, and the FIB would adjust its hash buckets accordingly. If
there was a way to do this with static routing (i.e. specifying a
traffic-share count), CEF would happily do this as well.

	oli


> ----- Original Message -----
> From: "Oliver Boehmer (oboehmer)" <oboehmer at cisco.com>
> To: "Alok" <alokdube at hotpop.com>; <cisco-nsp at puck.nether.net>
> Sent: Wednesday, September 14, 2005 9:35 PM
> Subject: RE: [c-nsp] Re: [j-nsp] load balancing between multiple BGP
> links 
> 
> 
> Alok <mailto:alokdube at hotpop.com> wrote on Wednesday, September 14,
> 2005 5:50 PM:
> 
>> Am not saying use it as a metric/make metric func(link_band) like in
>> EIGRP, just use instant value when scheduling/flow setup if there
>> are indeed equal cost paths to a destination
>> 
>> For example, it is typical to see ISPs back up high util link with
>> low util links between 2 critial peers, (or atleast I have seen it),
>> and I see no reason why I should simply not be allowed to use them
>> both 
>> why not use both?
>> Multilink is one option, this does not need that encap-mlppp either
>> ..right?
> 
> yes, but this would be only applicable to parallel links between the
> same pair of routers.
> 
> And the other option is BGP link bandwidth, as I (as well as a juniper
> guy) have suggested. This will allow to set up unequal-cost
> loadsharing 
> based on the ebgp link bandwidth.
> 
> oli