[c-nsp] BGP soft-reconfiguration inbound impact

[joshua sahala]

On Fri, Nov 27, 2009 at 9:11 AM, Mikael Abrahamsson <swmike at swm.pp.se> wrote:
> On Fri, 27 Nov 2009, Phil Mayers wrote:
>
>> It depends on how many routes you have I think. If you've got the full
>> feed, then I'd say you're going to pay a heavy price for soft-reconfig.
>
> Only if you modify the routes a lot via routemap or alike. This code has
> been much tweaked the past 5 years, so in some cases soft-reconfig inbound
> takes no extra memory at all.

memory isn't my problem currently...cpu utilisation on a 'clear ip bgp
$neighbour soft in' is

if you are heavily filtered and are relaxing the filters (increasing
the prefixes accepted), you should be fine.

if on the other hand, you are trying to restrict/depreference what you
are/were accepting, be prepared for a potentially massive cpu hit on
SR(B|C|D) (100% for 5-10+ mins - no *current* bug id, yet)...this of
course may cause other bgp sessions to flap, your igp adjacencies to
drop, and your 76xx swouter to be unusable.  admin'ing down the
neighbour you tried to soft clear may be your only resolution.

/joshua
-- 
A common mistake that people make when trying to design something
completely foolproof is to underestimate the ingenuity of complete
fools.
        - Douglas Adams -