At 8:18 PM +0000 7/18/03, Priscilla Oppenheimer wrote: >I think this would work, but why would you want to do it? Why not make those >two parts of Area 5 different areas? Simply from a documenation and human >communication point of view, you don't want the design to be confusing. When >someone refers to Area 5, you don't want someone else to have to say "which >Area 5?"
It's something I do frequently, given the assumption that an area corresponds to an application community of interest or a geographic region. >You'll notice that the OSPF RFC covers partitioned areas but only as >something that will work when an area becomes partitioned due to a network >problem. In other words, they don't consider it a good design practice, but >a workaround. Well, mostly for area 0.0.0.0 with virtual links. I have done some thoroughly ugly, but workable, nonzero area backups through area 0.0.0.0, with floating static routes running through backbone routers. > >What addressing will you use? OSPF does support discontigous subnets, so you >should be OK. However, avoid making this too complex and remember that it's >important to be able to summarize prefixes when injecting routes into Area 0. But maximum summarization isn't always the ideal, especially if it deals with blackhole routes. As soon as you summarize to any extent, you isolate area 0.0.0.0 from at least some flapping. If I have a block, say, /24 for an area, I often split it into "east" and "west" addressing blocks, and advertise /25s from each ABR. Unfortunately, there are two equally valid routing policies when you lose routes of an summary. I wish Cisco and Nortel would each add the method used by the other. Cisco, of course, will continue to advertise the summary regardless of the status of the more-specifics. This is good when you want maximum stability. Bay/Nortel RS, however, detects the loss of a more-specific, and will stop advertising the summary and start advertising the more-specifics. This is good when you want maximum connectivity. > >Design books always say to design OSPF hierarchically (and even go so far as >to say that OSPF requires a hierarchical design). But I think a partitioned >area is actually still allowed, just not a good idea? Comments, anyone else? >Thanks. Especially on Cisco, unless one is very careful on the level of summarization, partitioned areas can lead to significant blackholing. Is that desirable? It depends (on whether stability is most important). > >Priscilla > > > >alaerte Vidali wrote: >> >> Can you see any mistake in the following network? >> >> >> Rx ---area 5------R2----area 0-------R3------------- >> | | >> area 0 | >> | | >> Ry ---area 5------R1-------------------------area 0-- >> >> >> R1, R2 and R3 are connected through area 0. >> >> R1 and R2 are ABRs for area 5. >> >> I am wondering if R1 and R2 should be connected through area 5 >> for a better design. >> >> The bad situation I see is that Rx and Ry will have different >> databases, although they are in the same area. From the >> routing table standpoint there will be conectivity. >> >> Any Thoughts? Message Posted at: http://www.groupstudy.com/form/read.php?f=7&i=72633&t=72587 -------------------------------------------------- FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
