[c-nsp] Hierarchical FIB on Cisco 7600

[Oliver Boehmer (oboehmer)]

>Thank you all; so let me just see if I got this right.
>
>If we're not loadbalancing with IGP (instead there's primary/backup
>uplink)
>on edge, and not using H.FIB (with cef table output-chain build favor
>convergence-speed) and we're running full BGP table on edge routers,
>anyone
>with experience on how would flat FIB affect convergence without PIC Core
>on 7600? In other words what would be real life effect of rewriting couple
>of hundred thousand entries to point to new output interface in terms of
>convergence?

I don't have a recent performance study, but I saw a 7600 w/ 200k IPv4
prefixes taking ~17 seconds to rewrite all prefixes, where hierarchical
FIB cut this down to < 1 sec.

>
>If you don't mind, could you explain part with "VPNv4 requires recirc" ?
>I'm not sure what does that mean in relation with PIC Core?

it means it'll cut the pps performance in half for vpn traffic with H.FIB.

>
>And if I understand, since we're already installing in CEF 'repair' path
>in
>PIC Edge scenario, then H.FIB is not really required?

It is. We can't leverage the repair path unless we have a hierarchical FIB
structure.

	oli