[c-nsp] CEF-based per-packet load-sharing under MPLS VPN

[Oliver Boehmer (oboehmer)]

>> Have you considered a recursive load-sharing?
> 
> No, we haven't. I'm testing your suggestion
> right now, but no luck yet.
> 
> In my test scenario, interfaces Serial6/0/0:0 and
> Serial6/0/1:0 are attached to vrf 5500000001-99.
> 5500000001-98 is the other test vrf on the same
> PE. I have added these recursive routes for the
> test network 10.14.62.0/32:
> 
> ip route vrf 5500000001-99 1.1.1.1 255.255.255.255 Serial6/0/0:0
> 10.4.62.2 
> ip route vrf 5500000001-99 1.1.1.1 255.255.255.255 Serial6/0/1:0
> 10.6.3.130 
> ip route vrf 5500000001-99 10.14.62.0 255.255.255.255 1.1.1.1
> 
> This is what I see:
> 
> #sh ip bgp vpnv4 vrf 5500000001-98 10.14.62.0
> BGP routing table entry for 15180:4829:10.14.62.0/32, version 358152
> Paths: (1 available, best #1, table 5500000001-98)
>   Not advertised to any peer
>   Local, imported path from 15180:4824:10.14.62.0/32
>     0.0.0.0 (via 5500000001-99) from 0.0.0.0 (200.202.113.165)
>       Origin incomplete, metric 0, localpref 100, weight 32768,
>       valid, external, best Extended Community: RT:15180:1
>         RT:15180:4701 RT:15180:5001 RT:15180:5009

Hmm, the next-hop looks strange. I'd expect to see 1.1.1.1. Did you
import both prefixes (10.14.62.0 & 1.1.1.1) in the VRF? How does your
vrf config look like (or please send full config unicast)?

> #sh ip cef vrf 5500000001-98 10.14.62.0
> 0.0.0.0/0, version 0, epoch 0, attached, default route handler
> 0 packets, 0 bytes
>   Flow: AS 0, mask 0
>   via 0.0.0.0, 0 dependencies
>     valid null adjacency
> 
> Was this the expected behavior?

definitly not, you see a null adjacency here, possibly caused by a Null0
static default route in vrf 5500000001-98?

>> For dynamic PE-CE routing, iBGP/eiBGP multipath
>> should give you the knobs to import multiple
>> paths into your BGP table(s).
> 
> We'll consider dynamic PE-CE routing.

consider eBGP, then it will work much cleaner.. root-problem with two
equal cost static routes is that you will only see one BGP path after
redistribution. And since BGP is used to leak routes between VRFs, you
loose the information. With eBGP, you can use e/iBGP multipath and
announce all available paths..
   
	oli