VRF Routing Leaking and Performance

[Brad Nightingale]

Hi

I have a network in which I have several VRFs residing on one router (3925).  I need to be able to route between a few prefixes in the VRFs at a high rate, for traffic in on the same trunked interface (to go back out the same interface).  The router basically bottoms out at about 300M at 99% CPU.

I will probably replace this with a L3 switch (3850, which I understand has better performance than a 3650), as I would have thought that the line speed would be faster.

This is, of course assuming that VRF to VRF routing is CEF switched and NOT process switched.  Can anyone confirm this for me?

Thanks

[Peter Paluch]

Brad,

The traffic should be CEF-switched, at least to my quick experiments.

I wonder, though: How did you configure the leaking of routes between VRFs? Are you using BGP for this, or did you configure the routes statically? Would it be possible for you to share a portion of your configuration that relates to this?

Best regards,
Peter

[paul driver]

Hello
Enabling CEF on the L3 svi is desirable otherwise I guess the L3 interface would default to fast switching

You can check this by looking at your layer 3 interface ip stats, I think newer ios trains  have fast switching enabled by default but if I am correct older ios enabled process switching by default

Maybe someone could clarify this?

ip cef distributed

int vlan xxx
ip route-cache cef

sh ip int vlan xxx | in IP

res

Paul