Erblichs,
Thanks for the comments.

If you understand the basic philosophy behind the whole thing then most of
these questions are answered by themselves!

The understanding is that if an entity is flapping then
- you waste cycles running SPF with each flap which can in turn can result
in more adjacencies breaking and which may (in the worst case) lead to a
total network meltdown.
- you waste your bandwidth (no need to explain!)
- it (the flapping entity) must pay a price for this! It pays in a sense
because we don't use it as long as we gain enough confidence in it that it
is now going to function properly!

Now i will get on to your specific questions ..

> I need to think about this, but do you
> only hold back originated LSAs or
> transit LSAs into an area that only
> has a single edge rtr?

Currently we are only looking into the self originated ones and i *think*
we should only be doing this for the self originated ones or we can have
some scenarios of inconsistent database.

>
> If the route specified is the only
> route to a specified set of network
> segments, then what is the negative
> effect of advertising this flapping
> route?

you mean what happens currently .. right? we advertise a flapping entity.

well to begin with you waste your bandwidth, clog your processors/etc.
Things begin to get a little complicated when these IGP derived routes are
being used by BGP to recursively resolve the next-hops. Things can really
get very nasty if you are running BGP with thousands of routes and if it is
using those OSPF resolved next-hops. Moreover it can get worse if an ISP is
using its IGP costs to advertise as its MED!

>
> If the route isn't advertised, then
> the network segments are blackholed?

This is one price which a flapping entity *has* to pay! It is given enough
opportunities to correct itself (depends upon how the damping parameters
are configured!) and if it still doesnt do anything about it then yes ..
routes will not be advertised even though its up!

>
> If the route is advertised and the route
> is really unreachable, then the
> transit network segments are consuming
> wasted bandwidth. The packets will
> eventually be dropped anyway.
>
> If the route is advertised and the route
> flaps back positive, it will take some

what is "flaps back positive"? you mean it "flaps" ?

> time longer than 1/2 of a hello interval
> once a SPF algorithm removed the route,
> plus your penalty, plus a new SPF run,
> plus flooding time..

i couldnt get what you mean by this.

>
> Thus, if there isn't an alternative route
> THEN maybe just ignore the lost of the
> route... Stub areas and stub routes may
> not want to use the penalty.

we need to explore on this.

Regards,
Manav

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]