While we're on the subject of scalability, check out this fascinating
article by Geoff Huston on BGP scalability. He studied the effects of a
denser, more meshed edge of the Internet (due to small networks
increasingly multihoming) on the growth of the BGP routing table. He
comments that service resiliency in the Internet is becoming the
responsibility of the customer, not the service provider. Instead of
insisting on service provider resiliency, companies are implementing their
own resiliency by multihoming. This means that resiliency is provided
through the function of the BGP routing system rather than being a function
of the bearer or switching subsystem. Can the BGP routing system scale
adequately to continue to undertake this role?
Aside from Priscilla (not Geoff Huston): What if the phone system had
evolved this way? How many companies have redundant trunk lines? Don't we
just assume that the "phone company" will always provide service? We don't
multihome to the phone system, (do we?)
Back to Geoff Huston: Further driving the growth in the routing table is
the use of selective advertisement of smaller prefixes along different
paths in an effort to undertake traffic engineering. (MPLS)
Very interesting read. The URL is:
http://www.cisco.com/warp/public/759/ipj_4-1/ipj_4-1_bgp.html
Priscilla
At 08:38 AM 5/22/01, Howard C. Berkowitz wrote:
>See my earlier comments to Chuck.
>
> >So some people have said that IS-IS is more scalable because it doesn't
run
> >Dijsktra as much as OSPF does. OK, then why not? Is it because of the
> >partial-routing update thing, or is there more to it?
>
>True, both run Dijkstra, but ISIS allows longer timer settings of the
>equivalent to MaxAge. Thus, less frequent recomputations in a stable
>network, which is far more characteristic of a carrier than an
>enterprise network.
>
>And some carriers do run OSPF, for reasons ranging from feature
>support to simple familiarity. There are a lot of very poorly
>documented extensions to ISIS.
>
> >
> >Also, I agree that IS-IS level-1 areas are by their nature "totally
stubby".
> >But that doesn't completely explain why real-world IS-IS networks have
been
> >shown to be more scalable than real-world OSPF networks, because if this
was
> >the cause, then it would seem to me that you could just scale OSPF to the
> >same level of IS-IS just by making non-backbone areas totally stubby.
>
>I'm not exactly sure that the scalability argument is being made
>correctly. ISIS is definitely more scalable in single areas. Most
>providers run it as a single area. With the more extensive controls
>at the backbone to nonbackbone boundary, OSPF may very well be more
>scalable in more heterogeneous networks.
>
> > Yet,
> >apparently nobody has been able to scale OSPF like that, which indicates
to
> >me that it's not that simple.
> >
> >So I must ask again, what exactly is it about IS-IS that seems to make it
> >more scalable overall than OSPF? And, as a side question, could OSPF be
> >reasonably adjusted to accomodate greater scalability?
> >
> >I would especially like to invite Howard Berkowitz, aka Sir Network Deity,
> >to answer this question.
>
>In any case, there are many more factors of development interest than
>the number of routers per area. One carrier interest is getting IGP
>reconvergence time into the millisecond range, which will require, at
>least, much faster hellos, an update or replacement for Dijkstra, and
>probably restrictions on routes per area.
>
>Traffic engineering is also an issue -- the work is further along
>with ISIS, but that's simply been a matter of development priority.
>
>Some OSPF features that ISIS doesn't have include demand circuits,
>native NBMA support, external routers outside the backbone, virtual
>links, and the opaque LSA.
>FAQ, list archives, and subscription info:
>http://www.groupstudy.com/list/cisco.html
>Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
________________________
Priscilla Oppenheimer
http://www.priscilla.com
Message Posted at:
http://www.groupstudy.com/form/read.php?f=7&i=5468&t=5468
--------------------------------------------------
FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html
Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
