>Jack Walker wrote,
>Sorry for jumping into this.
>
>I think I agree with Bob that IS-IS is more like something a service
>provider should consider.
>OSPF is sufficient for enterprise network, at least I think so, just imagine
>a network with 3000 or more routers, how could we design a OSPF network
>like this? how many routers do we want to put into area 0, I never had a
>chance to work with a network at this level, but would like to know how it
>works.
One network I architected a few years ago had, at the time, about
2500 routers. At the time of implementation, most were Ciscos running
IGRP, but there were some Bay routers running OSPF. The need both
for scaling and multivendor compatibility pushed us to a
standards-based protocol rather than EIGRP. OSPF won over ISIS
because of greater familiarity and also better controls on
redistribution.
When my customer first said they had 2500 routers, however, a large
caveat emerged. Only about 400 routers in the organization had
active alternate paths (i.e., other than dial/ISDN backup). It was
only these routers that needed dynamic routing. 2100 or so edge
routers did perfectly well with static/default routing to
distribution tier routers, which used static redistribution into
OSPF. We needed a data base to manage addresses anyway, and this data
base generated the appropriate ip route configuration commands and
loaded them into configurations.
The customer had several major campuses with extremely stable campus
networks, and reliable WAN links (mostly DS3 with some multiple DS1
at cutover time). Off the top of my head, there initially were around
8-10 nonzero areas, one of which had most of the unstable links and a
small number of routers. Major sites had a pair of 7000's linked
with a LAN and each connected to 2 other sites, a backbone-only
router inside area 0.0.0.0 and to an ABR at another site.
Essentially, it was a ring with a star in the middle. Subsequently,
the customer put ATM switches at the concentration points, and
integrated voice and data in the ATM core.
Other large enterprise networks with which I dealt, however,
especially those that were intercontinental, tended more to be
regional OSPF domains (i.e., a regional area 0.0.0.0 and a set of
nonzero areas) interconnected with a "backbone of backbones" of
either static or BGP routes.
I have used ISIS for large provider backbone networks where there was
no real need for hierarchy, extremely reliable optical transport, and
the ability to have a single area. POPs feeding these networks might
run OSPF.
It all comes down to "it depends," and, if you are doing really big
networks, you need to know what you are doing. Lots of other factors
entered into all of these designs, such as the Internet connectivity
requirements.
>
>IS-IS definitely has the ability to handle a network at this scale, it is
>more robust than OSPF.
>I think the reason that not too many people like IS-IS is that enterprise
>or small service providers really do not need IS-IS, OSPF is sufficient. If
>you do not work with it, you do not know it well, and you do not like a
>thing that you do not know well.
>
>Just my 2 cents
>
>Thanks
>
>Jack
_________________________________
FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html
Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
