[c-nsp] Hierarchical FIB on Cisco 7600

[James Bensley]

On 23 April 2016 at 20:38, Adam Vitkovsky <Adam.Vitkovsky at gamma.co.uk> wrote:
>> Reviving this old thread, is anyone running a hierarchical FIB on a 7600, how is
>> it working for you? Has it halved your PPS rate for
>> VPNv4/VPNv6 traffic due to recirculation?
>>
> Sorry can't comment on this

I've had some off lists responses and spoken with TAC. For the sake of
the list archives, H-FIB on a 7600 (Cisco all it "PIC Core") will
indeed half the PPS rate for VPNv4 traffic, it's basically a no-go on
7600s, they aren't up to it. In addition to that, in this day and age
people are unlikely to have enough free RAM if they are carrying the
full BGP table to enable this feature.

An example 7600 in production I have shown to TAC with an
RSP720-3CXL-10GE has a sausage over 800MBs of free memory on RP and
just over 800MBs on SP; “show ip route summary” shows we are currently
using 128MBs of memory for 587,645  IPv4 routes (this router has the
full global BGP routing table NOT in a VRF). This will jump to 2KBs
per prefix, (587,645 * 2) / 1,000 = 1.175GBs of memory for FIB.



>> In the case that a PE connected to a transit provider goes bang the 7600s are
>> still slow to update. I don’t even have to measure this I can see it when
>> performing maintenance work on transit links (CPU shoots to 100% and SSH
>> slows down a bit etc whilst BGP churns).
>>
> To solve this you can use bgp pic-core then the speed of convergence would be reduced to how fast the IGP can propagate the BGP NH failure to ingress PE (that you can tune as well).


As above, it's basically not an option on the 7600s.


Cheers,
James.