Can somebody explain why software cef and hardware cef can results in different path traffic take?
And on my practice example - I have Cat6506 VSS system (Sup720-10g-3c) connected to two standalone Cat6509 (Sup720-3b) with MEC. Uplinks united into Portchannels occupied ports on Supervisors - so no DFCs envolved in forwarding traffic on VSS system (SW3).
The problem is that when i tried to investigate what path will take packet going from host located behind SW3 to devices located somewhere behind SW1 or SW2 i got different results with commands sh ip cef exact-route and show mls cef exact-route - in many cases software CEF and Hardware CEF choose different portchannels to forward traffic (from SW3 perspective) between same pairs on source and destination ip addresses. So i am a bit disappointed which path packets took really. I tried to find out it on practice using Netflow, but instead it made things more complex.
My understanding is this platform has RP and SP, which calculates the CEF load balancing independently. To avoid polarization, the hardware cef uses universal ID in additional to the source and destination IP to calculate the CEF load balancing; I think the universal ID is not considered when calculating cef load balancing in software.