Weird Netflow problem...

You don't configure it that way on a 6500. To configure netflow on the 6500 you need to configure "mls flow ip " The flow route-cache option doesn't capture CEF switched flows on the 6500 because they are switched by the MSFC card, so take those statements out.

Ok, I read up on the mls stuff, and I have added those configs.

But are you saying I should still remove all the ip route-cache flow entries from all my interfaces?

This is what my current config looks like now:

mls aging fast time 64 mls aging long 64 mls aging normal 64 mls flow ip full mls nde sender version 5 mls sampling time-based 64 ! interface FastEthernet7/1 description BGP Uplink ip address X.X.X.X 255.255.255.252 ip route-cache flow no ip mroute-cache ! interface FastEthernet7/2 description Server ip address X.X.X.X 255.255.255.0 ip route-cache flow no ip mroute-cache ! ip flow-export version 5 origin-as ip flow-export destination X.X.X.X 8787

Yes, you don't need them because the flows will be captured by the mls flows. The "ip route-cache flow" entries will not have all the information you need and will be duplicates of flows already captured via the mls flows.

That's not completely true. It's better to have both definitions in the config. Let me explain:

How MLS (Multi Layer Switching) and CEF works? First packet in the flow gets through Layer3 to find our a routing path. They Routing engine sends this information to Switching engine, and all remaining packets in the flow got switched instead of been routed. So, if you have the only "ip route-cache flow" statement, you will see the only "first packets", and if you have the only "mls flow", you will se the only "continuation traffic". So, it's better to have both.

Good luck,

Mike CCNP, CCDP, CCSP, Cisco Voice, MCSE W2K, MCSE+I, Security+, etc. CCIE R&S (in progress), CCIE Voice (in progress)

------ Headset Adapters for Cisco IP Phones

You are describing "fast switching", the predecessor to CEF. With CEF, all packets are hardware routed including the first packet. Fast switching built the FIB (forwarding information base) from the flows going through the router just as you describe. CEF builds the FIB from the routing table, so all packets are hardware routed including the first one.

Actually, what Mike described is "Netflow switching", which was the basis of earlier versions of MLS, such as the MLS used on the Catalyst

Fast switching is similar, but the cache is keyed on destination address, rather than on what one typically would call "flows". Although with MLS, it may have been possible to set the flow mask to "destination-only", so that Netflow switching would indeed devolve to fast switching.

Newer MLS implementations such as those on the Sup-2/PFC-2 and PFC-3 (Sup32, Sup720, RSP720) usually use (a TCAM-based implementation of) CEF for the forwarding decision, although in some cases (NAT, [IPv6] multicast?) they can do flow-based forwarding too.

PFC 3 / DFC3 has dedicated hardware for collecting netflow stats, based on a parallel set of TCAM to that used for routing.

i think the same happens on PFC2

only software switched flows have netflow collected by the MSFC