[c-nsp] Prefix Independent Convergence on 65/76k HW

[Saku Ytti]

On (2007-06-20 10:02 -0400), Tim Durack wrote:

> Have been reading the "Prefix Independent Convergence" presentation
> made recently at NANOG40. Looks very interesting.
> 
> Anybody know whether 65/76k hardware will be capable of supporting
> this down the line?

I believe 6500/7600 implement hierarchical FIB when configured with
MPLS FRR at cost of recirculation.

FOO#show mls cef mpls labels 722

Codes: + - Push label, - - Pop Label         * - Swap Label, E - exp1
Index  Local            Label                  Out i/f
       Label             Op
200    722 (EOS)        524293(*)              recirc
216    722              524291(*)              recirc
FOO#show mls cef mpls labels 524291

Codes: + - Push label, - - Pop Label         * - Swap Label, E - exp1
Index  Local            Label                  Out i/f
       Label             Op
98     524291           (-)                    Te1/1         , 0016.9c6c.xxxx


So label 722 does not point to Te1/1, but indirection/loadinfo (LI) label
'524291' which in turn is popped and packet is sent out via te1/1.
Now, if Te1/1 goes down, you don't need to update every label pointing to it,
you update just 524291.

You might want to try this:
BAR(config)#cef table output-chain build indirection ?
  non-recursive-prefix  non-recursive prefixes
  recursive-prefix      recursive prefixes

Or just do:
BAR(config)#cef table output-chain build favor convergence-speed 

I know that at least 6PE broke when setting this, and at that time I got
information that LI's (LoadInfos) are not supported in SRA, dunno if
this has changed in SRB. But sure enough, they still do it, want it
or not for MPLS FRR.

Thanks,
-- 
  ++ytti