[c-nsp] 6500/BGP/full route tables [even more] confusing...

[Jeff Kell]

This topic seems to come up periodically... and I thought I had it
figured out... but apparently not.  So I'm seeking some real-world
answers and opinions on my particular combination/options.

I had a 6509/Sup2 (clearly can't do full tables) for some time, using
various route-maps and filtering tricks to keep the IPv4 routes under
128K (which seems to be the magic number with uRPF enabled).  If that is
exceeded, it generates TCAM overflow errors and essentially the 6500 is
bricked relative to passing traffic.  Under 128K and it's fine.

I intended to upgrade this to "something" suitable for full routing
tables, and went for a Sup720/PFC3CXL.  A million routes, right?

Initial replacement resulted in the WS-X6516-GBIC / DFC blade being
refused and powered down.  OK, incompatibilty issue, but you can RMA a
suitable replacement. 

Received another WS-X6516-GBIC but with a DFC3A.  Powers up, but
switches everything to "PFC3A" mode:

> Router#show platform hardware capacity system
> System Resources
>   PFC operating mode: PFC3A
>   Supervisor redundancy mode: administratively sso, operationally sso
>   Switching resources: Module   Part number               Series     
> CEF mode
>                        3        WS-X6516-GBIC            
> CEF256          dCEF
>                        5        VS-S720-10G          
> supervisor           CEF

which means 192K unicast routes:

> *L3 Forwarding Resources
>              FIB TCAM usage:                     Total       
> Used       %Used
>                   72 bits (IPv4, MPLS, EoM)     196608         
> 25          1%
>                  144 bits (IP mcast, IPv6)       32768          
> 7          1%*

TAC says "If the routes are more than 192K, part of the unicast routes
will NOT be in the hardware and the traffic using those routes will be
punted to RP (Route Processor). This will result in high CPU."

OK, how much traffic would it take to choke a VS-S720-10G with RP
switching?   I'm not looking at pushing terabits here...

Official TAC recommendation is to get a DFC3CXL for the blade (do I
really want a $15K upgrade for an old 6516-GB blade?), or get a CFC
blade (I do have a WS-X6724-SFP / WS-F6700-CFC blade I could reluctantly
pull out of a core switch with some rearrangement), or get another
6724/CFC combination (interestingly enough, also $15K list).

Or I'm tempted to hang a 3750E off the Sup's 10G port and use it for gig
and to heck with the blade :)

Can't seem to get a straight answer relative to the RP switching
"penalty" here (an old Sup2 is handling the current traffic).

Thanks,

Jeff