>
> Chris Whyte wrote:
>
> > > Additionally, the question arises that if, say, there are
> 200 such
> > > interfaces on a given EDGE/ASBR, and one of them flaps...what is
> > > worse for the network? Producing a type-5 as-external-LSA for a
> > > single link state change (which propogates throughout the network
> > > essentially unchanged) or producing a type-1 intra-area-LSA which
> > > the ABR(s) still have to analyze in order to find the (1) network
> > > out of the 200 whose state has changed.
> > >
> >
> > Oh... I bet you're frustrated. 4 e-mails later and the
> question hasn't
> > even been addressesd by anyone, let alone me. Well, I feel
> obligated
> > to attempt now that I understand what you're asking.
>
> It was more a rhetorical than anything else, as it depends
> upon the implementation, but it's one worth asking since
> every network is different.
>
> > So, I imagine that it might vary depending on the implementation as
> > I've seen some ugly ones when it comes to not doing
> incrementals when
> > they could. If memory serves, neither option should require a full
> > Dijkstra at any router in the network since the flapping
> network is a
> > stub. However, I still think the latter is more beneficial
> since the
> > former has some, though possibly minor, impact on the *entire*
> > network. The impact of the latter can be limited to the
> local area as
> > long as it's being summarized at the abr. I've always been a fan of
> > isolating the visibility of any failure whenever possible. Plus the
> > other benefits just make it the cleanest solution in my mind.
>
> Again, this is presuming summarizations which may not exist
> in a legacy environment and may not be possible to implement
> in that environment.
>
> > In the many discussions I've had with developers in the past,
> > analyzing a router-lsa with lots of interfaces has never been
> > mentioned to me as an intensive process or just something to be
> > concerned about in general.
>
> I was searching for a comparative, and on the whole I am
> inclined to think that the impact could potentially be worse
> (however negligible it might be) in a legacy environment
> which it is not possible to summarize.
>
> If a person is running a 2501 with thousands of routes (that
> does happen) that person might be inclined to drop as many
> CPU cycles as possible.
>
> In the end, it's a toss-up depending upon your implementation.
Ok. Sounds good to me.
My only comment is that I've been directly involved in transitioning
more than one legacy environment, with a minimum of 300 routes and (low)
thousands of routes, towards summarization. I've managed to make it work
more than once, though it was quite painful. But I must admit, that
short-term pain had a tremendous positive impact on longer-term
stability. So it was well worth it, imho. I guess I can't completely
relate to the "not possible" comment. Kind of like Sprint's split-AS
design a few years back. Hell, you gotta do what you know gotta do.
However, I suspect the amount of relative pain would have to do with how
much of your customer's prefixes you allowed to get into your igp. And
we probably shouldn't spend any time discussing that one... :-)
Thanks for your time. I thought it was an interesting question once I
figured out what you were (rhetorically) asking.
Chris
>
> -J
>
>
This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:41 EDT