[c-nsp] Netflow Export Problem

[Phil Mayers]

Rubens Kuhl Jr. wrote:
> Packets originating from the router can be controlled by:
> ip local policy route-map <name of the route map>
> 
> One thing I would try, but I have no idea if it works, is policy
> routing the traffic to a loopback interface that belongs to a VRF.
> The netflow export address would be the one inside the VRF, the source
> will be inside the global routing table, but the PBR would match udp
> to that address and port and set the next-hop to the loopback
> interface.

I have my doubts that would work on hardware platforms, but anyway...

> Let's repeat: it's just an idea, not something from a reference or
> that had real testing.

I am fairly sure that won't work, for the simple reason that it won't 
always be the CPU/MSFC that generates the packets.

In many cases, the PFC or DFCs generate the netflow export packets. The 
Netflow/UDP packets would need to be emitted by the PFC/DFC directly 
onto the fabric bound for an output port(s), and thus a FIB lookup will 
need to take place. It's my understanding that the bit of the PFC/DFC 
doing the NDE doesn't lookup in the VPN/VRF TCAM. Whether it could be 
made to I don't know.

It might be possible to force the MSFC to handle the netflow packets 
(e.g. by enabling one of the options such as source/dest AS# which have 
to be "filled in" on the CPU) and then the VRF export might work; but it 
would cost MSFC CPU.