OSPF Network design

Wed Aug 15 21:16:42 UTC 2001

On Tue, 14 Aug 2001, jan Huizinga wrote:

> Hello,
> 
> I have a question regarding OSPF design.
> 
> I have a customer which has a hub and spoke topology, in the main site the
> have 5 routers and every remote site is using 3 routers and a 1/4 class C
> addresses. There are 8 remote sites.
> 
> They have created for every site an area, so in total 9 areas + an area 0
> makes 10 areas. Is this a typical OSPF design? At this moment this network
> is a 100% VoIP network (H.323). In the future they want also to give
> Internet access to their customers (dial-up and leased lines). The remote
> sites are connected with 1MB links, and some of them shall be upgraded to
> 2MB links.

There are two reasons that I can think of to have multiple areas within
OSPF.

1) Scalability: As the number of routers, and more importantly the number
of links in an area increase, scalability becomes an issue. Each link
state change requires a recalc of the SPF algorithm on all routers
participating in that area, and therefore multiple areas will both
decrease the amount of time to do the recalc, and also contain recalcs to
the area in which they occurred. Many people believe that you should begin
looking into multiple areas due to scalability when you are talking about
a number of routers in the tripple digits. Looking at your routers
statistics on time to recalc, and how often you recalc, will help you fine
tune when to make the jump.

2) Summarization: The other good technical reason you may want to break up
a network into multiple areas is to summarize at the area border routers.
In this way you could allocate out of blocks assigned within a specific
area and do route summarization at the boundary, making a bunch of /30s for
instance into a few /24s.  This could be because you want a clean routing
table, you are worried about scalability, or you just think it is cool to
do and you are anal about ip allocation.

There are a few other reasons to do multiple areas, but they mostly have
to do with organizational or administrative reasons. For instance, a
university might make each college or department run a separate area, but
connect them all with a common area 0. This could decrease the impact of
one college on the whole as area boundaries are the only place that you
should ever put OSPF filters, (a bad idea to do at all IMHO) and to limit
the scope of problems within a single department effecting the whole
campus.

There are some potential draw backs to having more then one area in OSPF.
The first is complexity. Unless there is a need for added complexity, KISS
should be the rule. Another reason is the topology. If the topology were
ever to change from a strictly hub and spoke network, the complexity
increases dramatically. For instance, if two of the spoke sites become
really important or have a bunch of traffic between them, you may want to
add a link between them to increase redundancy or reduce traffic on your
links to the hub. Or maybe you get a new hub site in the future. Having a
single area 0 makes this really easy. Virtual links to area 0 through
another area are a PAIN.

Knowing what little I do from your email, a single area 0 would fit the
bill nicely. 10 areas is probably way overkill in terms of complexity and
scalability with little or no benefit.

Eric

Eric Oosting               eric at netrail.net        office:404.739.4385
Sr. Network Engineer   Network Eng and Operations         NetRail, Inc

> 
> Any ideas? Or does some one have a good reference to a site or a book that
> deals with the designing of a (OSPF) network. I have books about OSPF but
> they talk all about the protocol, and don't give real world examples how to
> design this. 
> 
> Thanks,
> 
> Jan
> 