[c-nsp] OSPF LSA timers

[Raymond Lucas]

"Oliver Boehmer (oboehmer)" <oboehmer at cisco.com> wrote on 22/06/2009
01:47:27 a.m.:

> RE: [c-nsp] OSPF LSA timers
> Oliver Boehmer (oboehmer)
> to: Raymond Lucas, cisco-nsp
> 22/06/2009 01:47 a.m.
>
> Raymond Lucas <> wrote on Tuesday, June 02, 2009 08:10:
>
> > Hi,
> >
> > I have been gradually rolling out OSPF across a network including the
> > following bit of config:
> >
> > router ospf 172
> >  ispf
> >  timers throttle lsa all 10 100 5000
> >  timers lsa arrival 80
> >
> > Which was fine until I arrived at a couple of 6506s with SUP2/MSFC2
> > running 12.2(17d)SXB9 which don't support those commands.  Seems they
> were
> > only introduced in 12.2(18)SXF according to Software Advisor.
> >
> > We can't upgrade to 12.2(18)SXF due to a lack of memory on the switch
> > processors.  I'm not too worried by the "ispf" business, but I have a
> > bad feeling about having a couple of devices different from their
> > neighbours with the LSA stuff.  To really up the nerves, these 6506s
> > are are part of the core.  I can imagine it working well most of the
> > time but then failing badly when the pressure is on.
> >
> > So I guess my questions are:
> >
> > - Am I right to be worried, or will things work fine if I miss these
> > commands from these devices?
>
> It'll work most of the time, until you run into situation where you need
> to issue more than one LSA update per second (for the very same LSA id).
> As the other devices will ignore the 2nd LSA update, you'll have to
> retransmit and convergence will be delayed.

Yeah, did some testing and figured the same thing myself.  My local SE
provided similar feedback and also a hidden command that would stop these
devices ignoring those 2nd LSA updates.  I would rather not use hidden
commands if I can help it though.

> > - Since these timers can only be set on a per device basis, as
> > opposed to per interface, is there an elegant way to deal with this
> > scenario? Obviously I would not be keen to remove the modified timers
> > from the rest of the network!
>
> Well, if you really need these timers to meet your convergence targets,
> you're out of luck and need to upgrade the devices. I would argue,
> however, that you'll be able to get away with less aggressive LSA update
> timers (i.e. timers throttle lsa update 10 1000 5000) in most scenarios.
> Tuning SPF timers is usually more important..

Nice to know.

In this particular scenario, it is currently not so important to have quick
convergence in Area 0 where we have this problem.  We don't have the
problem in other areas which is important to us.

Thanks heaps for the feedback,

Ray