[nsp] BGP tweaking possibility

[Eric Osborne]

On Wed, Jul 16, 2003 at 01:50:13PM -0400, Tony Tauber wrote:
> On Wed, 16 Jul 2003, Gerald wrote:
> 
> > > Very true.  There's no mechanism to cut traffic up by volume.  The
> > > best you'll be able to do is trial and error unless you get into
> > > some kinky netflow stuff.  Even then, you're making a config based
> > > on some statistics, not causing the router to make a decision
> > > based on load.
> >
> > That's kind of the automated version of what we are currently doing.
> > MRTG + modifying route prefs & prepends.
> 
> Just to clarify, what I was talking about was not automated.
> It's possible with Netflow to pull out info about traffic flows
> broken down by AS number.  You still have to decide what config might
> get you what you want through local-pref and prepends to pack the
> traffic "chunks" to achieve the desired effect.
> 
> > > You might do better to take both transit and non-transit routes
> > > and depreference the transit ones so they're only used in case of
> > > a failure of the preferred path.
> >
> > My understanding of that is if one of the ISP's changes their
> > network slightly after I put that in place, that could throw off my
> > fine tuning.  I'm hoping to make it an internal router decision
> > based on bandwidth.
> 
> That's right.  I don't know of a way on Cisco (which is not to say
> there isn't one) to make a decision based on bandwidth.
> 

http://www.cisco.com/en/US/partner/products/sw/iosswrel/ps1829/products_feature_guide09186a008012db46.html
(BGP Link Bandwidth)

might help in some (although certainly not all) situations.

http://www.cisco.com/en/US/partner/products/sw/iosswrel/ps1829/products_feature_guide09186a008012ee22.html
(BGP Cost Community)

may also help, depending on whether you want load-sharing or a
cost-based tie-breaker.



eric



> > I've had 2 good suggestions off list so far. One is a commercial
> > solution (ick) ...and the other I don't fully understand yet so I
> > have to go do more Cisco reading. The recurring cisco wording from
> > the other suggestion on and off list is to use "Policies" to do what
> > I want. /me grabs Cisco dictionaries/URLs to read up on what
> > policies will do.
> 
> My understanding of the RouteScience and Sockeye boxes is that they
> can affect the traffic outbound from you to your provider(s) by
> adjusting the routes they hear on the basis of some criteria and
> perhaps bandwidth is one of them.  If the other suggestion is BGP
> policy propagation, from the use I'm familiar with, it doesn't do what
> you want.
> 
> > Anyone actually done this that could e-mail me off or on list?
> >
> > > You could act on the communities on routes that you're hearing
> > > from others, which would affect outbound traffic.  Some providers
> > > will act on communities you send to them in order to change the
> > > preference of a route relatively up or down somewhat which will
> > > affect inbound traffic.
> >
> > Like AS-Prepends? Already doing that and not getting the results I
> > want yet.
> 
> Right, but instead of you prepending all the advertisements, they can
> prepend to selected providers to give more granular control.
> 
> Tony
> 
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> http://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/