First, don't misunderstand me. ISIS and OSPF are both good protocols,
and the decision to use one or the other in a given provider network
can come down to personal preference. OSPF has a lot of features
that tend to be needed more in an enterprise, but can be
bandwidth-conserving from a POP to a core when bandwidth is an issue.
A lot of the Cisco "traditional wisdom" about OSPF is meant as
guidelines that will work in the worst case. The restriction on
number of routers per area assumes routers with fairly slow CPUs, and
fairly unstable links. You could put a lot more 7200's than 2500's in
an area!
The number of areas per ABR is based on the statistical likelihood of
there needing to be a simultaneous recomputation in more than one
area. I've had no trouble hooking seven areas to an ABR, where each
area was an all-optical, well-maintained campus.
>Use the search string "isis" ( no dash )
>
>From my own limited studies:
>
>IS-IS tends to treat level-1 areas as stub networks - therefore smaller
>routing tables
Standard ISIS doesn't just treat them as stub, but as totally stubby.
While the majority of ISP implementations are a single area anyway,
this restriction has become a limitation for some networks, and
controlled L2 leaking has been added to the protocol. Sort of the
reverse of OSPF, where you primarily decide in the nonbackbone area
what to advertise in the backbone.
>
>IS-IS sure looks a lot chattier than OSPF. Debug ISIS adjacency reveals a
>LOT of traffic generated just by the protocol keepalives
In principle, yes. They are lighter weight, and the timer range is
greater than in OSPF.
>
>The killer in OSPF is the SPF algorithm, and the related processor intensive
>activity It has less to do with size of routing tables.
Not really. Both use the SPF algorithm, although a little more
detail is needed to understand this. Intra-area routes in both use
the Dijkstra algorithm, whose processor load is on the order of:
(number of routes**2) * log(number of routers)
Inter-area and external routes are in the second part of the SPF
algorithm, and their additional load is linear with the number of
routes. ISIS, of course, is far less likely to have lots of
inter-area and external routes.
>
>IS-IS tends to be more efficient than Ospf in that it is not beholden to the
>area 0 concept and the necessity for all inter area traffic to go through
>area 0.
Ummmm...well, ISIS doesn't have an area 0.0.0.0 per se, but it does
have a backbone: the set of L2 routers. All inter-area traffic
still has to go through the backbone.
>
>BTW, I have been told by folks who work in really big networks that none of
>the routing protocols scale beyond 4-5K routers.
I'd tend to agree, although I might not even put the limit that high;
more like 2-3K. In really big networks, there usually are good
reasons why you want to separate things into routing domains and have
a backbone-of-backbones. For example, I dealt with a couple of
intercontinental networks where the North American, Asian, and
European line speeds tended to be significantly different (DS3,lots
of 64 Kbps and some T1, E1 and lots of nailed-up ISDN), and it was
far more deterministic to have fairly consistent line speeds at each
level of the hierarchy.
Message Posted at:
http://www.groupstudy.com/form/read.php?f=7&i=5419&t=5207
--------------------------------------------------
FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html
Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]