[j-nsp] Summarize Global Table

[Chris Morrow]

On 10/26/2011 09:56 AM, Pavel Lunin wrote:
> 
>> The removal of a large aggregate (say 4/8) and the subsequent required
>> addition of all of the subnets under 4/8 could certainly be
>> 'interesting' to observe.
>> Interesting as well is that 4/8's origin doesn't have to be the flapper,
>> just your network (or an intermediate) changing attributes (or exit
>> path!) could be enough to induce this problem, no?
> Right! This's why it's not bad to play with it around and see whether
> this simple idea really works as expected in real-life.
> 
> On one hand 4/8 withdrawal or change should not be worse than installing
> some part of the full table. Very-very roughly, /8 is about 1/200 of the
> full table, which is ~2000 prefixes by now (can be checked but I'm too

keep in mind that the +/- of 2k prefixes is IN ADDITION to all the other
work going on in the routing table... so how many (and how often) +/- 2k
hits can you take?

> lazy). Not so few but looks acceptable in case a /8 flaps. On the other
> hand, I personally wouldn't try to predict, how often this can happen
> with longer aggregate prefixes for the full table overall, and whether
> it's acceptable. If someone thinks it's obvious — I'd like to hear the
> explanation.

the data (some of it at least) is available in MRT dumps from
ris/routeviews, right? so some modeling of this could be done
quickly/easily.

Now, the fun part... in your internal aggregates, if those change,
what's the impact? or do you have to keep all customer prefixes in your
local table all the time? (and in va-terms not mark them for suppression).

-chris