At 11:50 PM +0000 7/18/03, Priscilla Oppenheimer wrote: >alaerte Vidali wrote: >> >> Thanks. >> >> The addresses are contiguous. >> >> Suppose a network with many ABRs, one in each city. Any big >> city represents small cities. Could you use an area for each >> ABR? (I am wondering if there is no limit in the number of >> areas. I bet not). > >Yes, you should use an area for each ABR. Well two areas per ABR, actually. >As you probably know, OSPF networks have a two-layer hierarchy: the backbone >and the attached areas. The topology is a logical star of areas. Think of a >daisy with a center circle and petals surrounding it. Each ABR connects to >Area 0 (the backbone) and at least one other area. (An ABR could connect to >a couple other areas too, if that fits your needs. But an area shouldn't >span 2 ABRs, which I think was what you were suggesting.)
Now I'm unclear. Having ABR1 and ABR2, each with an interface into areas 0.0.0.1, 0.0.0.2, and 0.0.0.3 works perfectly well. You just have to be careful about partitioning and summarization. > >Areas are supposed to be contiguous (all in one connected piece). If links >fail and render areas other than area 0 discontiguous, OSPF can handle this, >but it's not how you are supposed to design it. Yep...so it means that you don't design an area with a single point of failure link, or go creative with such things as demand circuits. If an area represents an application community of interest, it's to your advantage to keep traffic intra-area. > >Over the years, I have collected the following design recommendations which >may be a bit old, and of course, the real answer is "it depends on router >CPU and RAM, what else the router is doing, how stable routers and networks >are, topological requirements, performance requirements, etc." > >* An area should contain no more than about 50 to 100 routers. >* An OSPF autonomous system should contain no more than about 100 areas. >* A router should be in at most about 3 areas. It's a question of stability of the area. I've had it work nicely with 5-7. >* A router should have fewer than about 60 adjacent neighbors. The number of routers really reflects non-RISC CPUs. > > >You could put the ABR in each large city and have each area represent a big >city and the small, neighboring cities. Then your backbone could be the WAN >between cities. > >A lot of experts recommend keeping the backbone small, fast, easy to manage, >reliable, etc. however. And a large inter-city WAN might not meet those >needs. So instead, make the backbone a set of routers at a central site >connected via Gig Ethernet. If there's a good place to have a centralized site, or perhaps two that are multiply linked, this is a good idea. In those cases, I try my best to have the WAN links stay outside the backbone. If the backbone is naturally WAN connected, as in a set of campuses, I usually set it up as a ring, with each ABR having paths to two other ABRs and/or a hub backbone router. When possible, if reliability is an issue, each "ABR" is actually a pair of physical routers (or a fully redundant router) with a high-speed link between them. I also like to use parallel links between the pairs of ABRs, which again adds cost, and makes no sense unless you have at least interface diversity. > >Another alternative is just to have one area, Area 0. Are you sure you need >multiple areas? NOOOOOO! If you have only one area, number it ANYTHING except 0.0.0.0. That way, if you need more areas, you only need to start renumbering in the backbone and the new area(s). > > >> What about put toghether two cities and form a bigger area with >> two ABRs? > >Why? There's no reason to do this, is there? > >Here's a paper by Peter Welcher that might help: > >http://www.netcraftsmen.net/welcher/papers/ospf1.htm > >Also check Cisco's OSPF design guide, although the current versions has a >tendency to say what you can do rather than what you should do and refuses >to give any definte recommendations, saying instead things like, "For this >reason, it's difficult to specify a maximum number of routers per area. >Consult your local sales or system engineer for specific network design >help." :-) > >http://www.cisco.com/warp/public/104/1.html > >Also check papers and books by Howard Berkowitz. In addition to 2 OSPF >papers at CertificationZone, he also has a paper on scaling routing protocols. > >Priscilla > Also see a NANOG presentation at http://www.nanog.org/mtg-9910/ospf.html Message Posted at: http://www.groupstudy.com/form/read.php?f=7&i=72637&t=72587 -------------------------------------------------- FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
