[c-nsp] OSPF design

[Keegan Holley]

On Fri, Oct 22, 2010 at 2:22 PM, Geoffrey Pendery <geoff at pendery.net> wrote:

> > Not true.  All /prefixes/ will still be visible in all areas without
> summarization, but you'll still minimize the LSDB because type 1 and 2 LSAs
> never cross an area boundary.
>
> Right, but as Brett mentions, that doesn't stop them, it just turns
> them into type 3s.  Sometimes it even turns a single type 1 (with stub
> networks on it) to multiple type 3s.
> I am curious if someone can explain any benefits to non-summarized
> areas.


The main benefit is that link and router failures in separate areas don't
trigger SPF/Diijkstra.  Also, when SPF is run it's not as hard to calculate
a path to a destination outside of the area.  As someone mentioned these
guidelines are rather old, definitely before routers with 1G or more of RAM
and multi-core processors.


>  Perhaps the flooding model is reduced a bit by relying on a
> small number of ABRs to flood the type 3s, rather than intra-area type
> 1 and 2 flooding?
>

Flooding is still relatively the same.  The number of LSA's remains the same
they are just of a different type.  So your neighbors will still send you
the same number of routes even if they are advertised from one or two ABRs.

>
> I honestly don't know, and would like to.
>
> > According to Jeff Doyle, control plane performance concerns are not
> related to  running Dijkstra, but rather lie in LSDB maintenance:  Keeping
> track of timers and such.
>
> So if this is the case, it would seem the LSDB doesn't get any smaller
> (with type 3's subbing in for 1 and 2), except now ABRs have to
> maintain multiple LSDBs, thus doubling/tripling their burden?
>

I'm actually not sure about this either.  Specifically, why it's easier to
manage an LSDB full of type-3's rather than one full of type1/2's.  Also,
does any of this still matter?  Many of the higher end routers have as much
processing power as servers did when these standards were written.

Keegan