I am a great fan of using separate OSPF domains, with the area
0.0.0.0's linked by a backbone of backbones. Most commonly, that
backbone will use BGP, for exactly the reasons of control that Alan
Robertson points out. I have had success, in specific networks, of
building that backbone with static routes (including floating
statics). These days, there are even arguments for using MPLS in the
backbone as well.
A Static Approach
-----------------
One network in which I used statics had an Asia-Pacific, North
American, and European region, plus a substantial corporate HQ in the
US. There was a good deal of internal communications in each region,
and substantial communciations with corporate, but not that much
region-to-region communications. The Internet firewall was at
corporate. It was reasonable to do the small amount of inter-region
communications via the corporate center.
Each region characteristically had different speeds in its backbones,
DS3 in North America, E1 in Europe, and 64 Kbps in Asia. The
connectivity back to corporate used international frame relay with
international ISDN backup.
At corporate, we set up a collapsed backbone LAN which had two
routers for international use, each with a serial and a PRI interface
so we could have diversity -- the backup ISDN link never went to the
same physical router as the frame link for the same region. I did set
up a small OSPF domain at headquarters, which mostly was for
separation of corporate server farms from corporate user/local
traffic, but also gave us a framework for connecting the networks of
mergers and acquisitions.
A BGP Backbone
--------------
For a different worldwide customer, I chose a BGP backbone.
Initially, each region was a single OSPF domain that appeared as a
BGP confederation to the backbone. This company was in the
shipping/logistics business, and, while some major data centers were
in the US, there was very substantial inter-region communications,
and communications from each region to the Internet.
BGP let us control some fairly complex backup arrangements involving
transoceanic bandwidth. It also facilitated the complex,
multiprovider Internet connection strategy.
>Peter,
>
>OSPF has a distance of 110, and yes, iBGP has a distance of 200. By
>having seperate routing domains for North America and Europe, he could
>use eBGP (Distance - 20) between his two networks.
>
>Distance wouldn't really do anything in this case, though, because
>European routes would not be learned via OSPF (Remember, we have
>theoretically split OSPF into two seperate routing domains, never the
>two shall meet).
>
>Instead, eBGP would be bridging the gap between the two OSPF networks.
>This would afford the opportunity to really take control of what
>routes were advertised between the two, and excercise strict control
>of the routing metrics, manipulating them in such a way as to ensure
>that the best path across the pond were utilized under all normal
>circumstances, but providing the redundant "less preferred" path in
>the event of some kind of outage.
There might also be some very complex issues if convergence and
stability needed to be considered. I can control flaps better with
BGP, but OSPF converges faster. Of course, if I'm only using a few
routes, the difference may come out in the wash.
>
>Can the same be accomplished via OSPF? Yes, but because we're dealing
>with Intra-area, and Inter-area routes, it may be more complex than by
>simply manipulating the link costs. Remember that OSPF chooses an
>Intra-area route with a Cost of 4,000,000 over an Inter-area route
>with a cost of 100. That's just one of the quirks of the protocol.
>
>As for "Why would you want to break up an AS that small into two
>seperate private ASes?", it's called thinking outside the box.
A routing domain is a set of addresses and routers with a common
routing policy, common IGP and metric assumptions, and common
administration.
An AS is a set of addresses and routers, under one or more
administrations, that presents a common routing policy to "an"
(lower-case) internet, which could either be the public Internet or a
complex intranet or extranet. If the policies of the domains are
significantly different, then separation into AS makes sense.
>We
>tend to think that a small network could not be better served by
>applying the same principles that we might use for a larger
>environment. Why is that? Instead of letting the number of devices
>determine the right solution (Or more properly, a good solution),
>let's form a solution based on the specific requirements.
>
>A network with a small number of devices, but consisting of multiple
>sites, and redundant links, presents a unique challenge. Forget the
>number of devices, and look at both the physical topology, and the
>problem that needs solving. BGPs powerful policy routing tools make
>it a good fit for this environment, when viewed from a requirements
>perspective.
>
>It's not the only solution, but it is a valid solution, and in my
>opinion, it's a good solution.
>
>Alan
>
>----- Original Message -----
>From: "Peter I. Slow, CCNP Voice Specialist"
>To: "W. Alan Robertson" ;
>
>Sent: Tuesday, May 29, 2001 1:02 PM
>Subject: Re: Wanna Be a CCIE? Try This One [7:6076]
>
>
>> Absolutely, but he has traffic going from one router to another,
>it's not
>> ever exiting the system.
>> ...why would you want to break up an AS that small into two seperate
>private
>> ASes?
>> besides... the OSPF routes are going to take precedence, not that
>the admin
>> dist. cant be changed, but ospf is 120, and BGP int routes are
>200....
> > (right?)
> >
Preference can be a factor, but summarization/aggregation is even
more important. More-specifics are always preferred.
> > ----- Original Message -----
>> From: "W. Alan Robertson"
>> To: "Peter I. Slow, CCNP Voice Specialist" ;
>>
>> Sent: Tuesday, May 29, 2001 12:42 PM
>> Subject: Re: Wanna Be a CCIE? Try This One [7:6076]
>>
>>
>> > Peter,
>> >
>> > With all due respect, he doesn't have an IGP problem... He has a
>> > routing problem, and would like the ability to influence the flow
>of
>> > traffic under certain circumstances to provide for better network
>> > performance.
>> >
>> > After hearing a better explanation of the real issue, path
>selection
>> > for an International site, the use of BGP might go a long way
>toward
>> > solving the issue.
>> >
>> > He could very simply address his issues by breaking his OSPF into
>two
>> > seperate routing domains, and utilizing BGP as a means of
>> > interconnecting them. He could manipulate the traffic through the
>use
>> > of something as simple as AS-path prepending, or the other
>mechanisms
>> > Chuck mentioned (local preference, weight, or meds).
>> >
>> > Routing protocols are but tools, a simple means to an end. Like
>all
>> > tools, each has it's strengths and weaknesses. Most important is
>that
>> > you select the right one for a given situation. In the absence of
>> > more information, the use of BGP sounds like a pretty good
>solution to
>> > the given problem.
>> >
>> > Alan
>> >
>> > ----- Original Message -----
>> > From: "Peter I. Slow, CCNP Voice Specialist"
>
>> > To:
>> > Sent: Tuesday, May 29, 2001 11:29 AM
>> > Subject: Re: Wanna Be a CCIE? Try This One [7:6076]
>> >
>> >
>> > > next time you recomend using bgp to fix an IGP problem, im going
>> > to.., well,
> > > > uh, just dont do it again.
Message Posted at:
http://www.groupstudy.com/form/read.php?f=7&i=6274&t=6274
--------------------------------------------------
FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html
Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
