[j-nsp] BGP advertise-inactive (even not the *best* BGP route)

[Pedro Roque Marques]

Tony Li writes:

> Actually, the hack would be to stick a route identifier into the
> mix, just as was done in IDRP.

You mean like in inet-vpn (2547) ? ;-)

If what the original poster wants is to be able to carry different
views of routing through the network bgp inet-vpn can do the trick
quite effectivly...

One example:
suppose that you want to load-balance traffic between a couple of
private peerings for most of your traffic going out to a provider but
you really want to also have the option of doing closes exit for a set
of customers/traffic.

In the gateways that connect to the remote network, configure:

routing-instance closest-exit {
    instance-type vrf;
    vrf-target target:65000:1;
}

protocols bgp {
    group ibgp {
        family inet unicast;
        family inet-vpn unicast;
    }
    group peer {
        family inet unicast {
               rib-group rg-select;
        }
    }
}

routing-options {
    rib-group rg-select {
        import-rib [inet.0 closest-exit.0];
        import-policy po-select;
    }
}

/*
 * apply import policy into inet.0 to reduce the local-pref on a given
 * subset of prefixes that will have higher local pref from other exit
 * point. apply change to inet.0 only.
 */
policy-statement po-select {
    term a {    /* can also be done via bgp import policy */
         from prefix-list foo;
         to rib inet.0;
         then local-pref low;
    }
    term b {
         to rib closest-exit.0;
         then local-pref 100;
    }
}

then in your ingress point you can configure

routing-instances closest-exit {
    instance-type vrf;
    vrf-target target:65000:1;
    routing-options static route 0/0 next-table inet.0;
}

And configure FBF so that packets matching whatever criteria you
choose get forwarded via the closest-exit instance... if the route is
not on the table then it will default to use inet.0.

You can build all sort of applications that require different views of
routing via FBF + 2547...

2547 is not tied to VPNs... it just adds the support for a route
indentifier (called route distinguisher) and a policy mecanism that
dictates which routes get imported into which tables (route-targerts).

All the stuff you guys are asking for already exists...

  Pedro.