[c-nsp] OSPF, load balancing

[Rodney Dunn]

There are all kinds of hacks we could probably come up with and
some are uglier than others.

It all depends on what your tolerance is. :)

You coul do NAT on the outside interface and then only route traffic
back to that ip address over the upper path for example.

Similar to what is done for dual ISP connections. The shortest IGP
path back to the address you NAT to would force it to be symmetrical.

No easy way wit OSPF in an equal path scenario to direct the traffic
one way vs. the other.

Let me ask the OER folks. I thought they were working on something
at one point that would actually fix this.

Rodney

 

On Wed, Mar 07, 2007 at 03:06:38PM +0100, Paolo Losi wrote:
> Rodney Dunn wrote:
> > Short answer: No. You can't get symmetrical
> > flows like this. The CEF hashing algorithm will hash uniquely
> > at Router A and Router B and while deterministic at the local
> > router it's nondeterministic at both points in regards to a flow
> > path.
> > 
> > Why do you care which way it goes unless there is some stateful firewall
> > or feature enabled on them?
> 
> 
> That is exactly the reason. We do stateful traffic shaping on those boxes.
> It would be acceptable to "manually" divide the traffic (e.g.
> by /24). We could use static route and route-maps with source routing
> but, at least, we would like to invalidate those routes/ route maps
> according to OSPF next hop availability...
> 
> Is this feasible?
> 
> Thanks again!
> Paolo