why would I use the command "ip ospf network point-to-point" on broadcast links and not have DR and BDR election?
I know the links would form adjacencey with each other but... any other reason??
If you have a network link that is a broadcast link and if it has only 2 devices on it that can run OSPF then you could configure the OSPF network type as point to point. In doing this you would eliminate a (small) amount of overhead that the OSPF process uses to negotiate DR/BDR relationship and to track DR/BDR state.
For example if you removed the VLAN from the switch connecting them? That's an interesting point. Both routers would detect the neighbor down, but they would both continue to advertise the network to the rest of the topology. But I don't think there would be any other change from the broadcast situation.
This does work, as you have observed. The only reason I can think why you should do it is that the CCIE scenario told you to do so!
I suppose you avoid the DR/BDR election, so it will come up slightly quicker. What would happen if you added third router in the mix could be anyone's guess.
If you did configure this and when there is a network change it would work just the same way that it works if routers are connected by point to point serial links. The routers have formed neighbor relationships, they track each other, and when they have a change they send the LSA to the neighbor.
As Kevin points out one reason that this is not commonly configured is the possibility that the broadcast link could get a third OSPF speaking device. In that case the DR/BDR really do help the network scale better.
And the savings in overhead of not having the election is very slight. So why take the chance?
Kevin, nothing to do with CCIE lab scenario.
This is a large network and all broadcast links use the point-to-point command....so no DR/BDR.
Sorry, I was just joking there. ;-)
If it is a large network, I guess they were just trying to minimise the convergence time. Did they also adjust the timers?
It is interesting that this is not a CCIE or lab question but a question about a live network. As I commented the DR/BDR do help networks scale better when there are more than 2 OSPF speakers on the link. If every router has to actively neighbor with every other router on the link it gets quite complicated as the number of neighbors increases.
I don't have access to my lab from here, so I cannot try it out. Do you know what happens if you do a point-to-point on a multiaccess network, and a third router (also p2p) joins in? Can a point-to-point interface hold more than one neighbor? Would you get a full mesh?
I do not know for sure what happens with a third router on the link, assuming that all routers on the link are configured as point to point. I believe that all routers form neighbor relationships with each other router on the segment.
I believe that an interface configured as point to point network type can have more than one neighbor. I believe that you do get full mesh. This results in a sub-optimal situation. With full mesh each router must maintain the neighbor relationship and the data base sync directly with each other router on the link. With full mesh the number of peers is calculated as n times (n-1) and n square relationships do not scale well.
All 7600s and 6500s have neighbor relationship on every interface with the P2P command..but it only sees the connected interface as a neighbor.
Are there any of these broadcast links that have 3 OSPF speaking devices on the link?
With 2 OSPF speakers I believe that there is not any particular functional difference, and you have saved a small amount of overhead. With 3 PSPF speakers I believe that it is slightly negative in terms of efficiency. As the number of neighbors increases the loss of efficiency increases rapidly.
In trying to understand what is going on here it may be helpful to clarify our understanding of becoming a neighbor and of establishing adjacency. Some people regard them as the same but they are actually different. Just because 2 routers have become neighbors does not mean that they have necessarily established adjacency.
In becoming a neighbor routers send and receive hello messages. They check parameters on the hello message to determine if they should become neighbors (do the timers match, are they on a common subnet, does the area type match, etc). If all the parameters match properly then the routers become neighbors.
To establish adjacency the routers sync their data bases going through the DBDs and the LSAs to verify that they each have the same understanding of the network topology.
For point to point links (which have no DR/BDR) neighbors automatically proceed to form adjacency. On multi-access links the routers must consider DR and BDR in determining whether to form adjacency. Two routers will form adjacency only if one of them is DR or is BDR.
So point to point is slight more efficient when there are only 2 OSPF speakers. And multi-access/broadcast with election of DR/BDR is more efficient if there are 3 or more OSPF speakers.
This router has more than 3 speaking OSPF neighbors, here's a sh ip ospf neighbnor output:
RTR1#sh ip ospf neighbor
Neighbor ID Pri State Dead Time Address Interface
xxx.xx.1.4 0 FULL/ - 00:00:35 xxx.xx.130.6 GigabitEthernet1/24
xxx.xx.1.2 0 FULL/ - 00:00:35 xxx.xx.130.14 GigabitEthernet5/1
xxx.xx.1.3 0 FULL/ - 00:00:30 xxx.xx.130.2 GigabitEthernet5/2
xxx.xx.1.3 0 FULL/ - 00:00:32 xxx.xx.131.10 GigabitEthernet1/3
xxx.xx.4.161 0 FULL/ - 00:00:32 xxx.xx.131.1 GigabitEthernet1/1
xxx.xx.4.165 0 FULL/ - 00:00:38 xxx.xx.131.5 GigabitEthernet1/2
xxx.xx.4.169 0 FULL/ - 00:00:32 xxx.xx.133.21 GigabitEthernet1/4
xxx.xx.4.173 0 FULL/ - 00:00:34 xxx.xx.133.25 GigabitEthernet1/5
xxx.xx.2.17 0 FULL/ - 00:00:37 xxx.xx.0.6 GigabitEthernet1/7
xxx.xx.2.13 0 FULL/ - 00:00:34 xxx.xx.0.2 GigabitEthernet1/6
Wow! I suppose all the other routers on the VLAN must have something similar. That makes 55 adjacencies going on just for this single Ethernet. Doesn't that have an negative impact on the SPF calculation time, as well as the LSA flooding?
Ah! I think I've just fallen in! These 10 neighbors are on L3 interfaces. I had thought they were on the same VLAN. Of course, if they were on the same broadcast segment, the interface would say VLAN n, and not GigbitEthernet ... . This is a layer-3 distribution switch.
Do you have any Ethernet segments where there are more than 2 routers? If not, then point-to-point is fine. Effectively you are using each Ethernet segment as a P2P link.
Have a nice weekend!
I just tried to put three routers'fa0/0 on one broadcast segment (a L2 switch) and put all of them on point-to-point ospf network type. It threw up all sorts of logs and won't form an adjacency. It seems it will accept no-nonsense on a point-to-point link and just hold one neighbor on interfaces with type set to p2p.
Thanks Arav, with this sort of question it all boils down to "how does a real router behave?". It's great that on NetPro there are so many people who are passionate about the subject that they will lab up the scenarios. Thanks.
So, looking at the original posting, I think we have established that these were layer-3 distribution links. If they were no switchport to no switchport, then they might as well be treated as point-to-point.
There is one thing to say about (real) point-to-point links: they detect link failures at both ends of the link simultaneously, whereas an Ethernet network does not - it has to wait for the hold time on the OSPF or the BFD to expire. That has nothing to do with the OSPF logical network type, but more to do with the nature of the layer-1 connection. In this respect, a no switchport to no switchport connection behaves more similarly to a P2P serial link than to an Ethernet. Maybe that is the rationale behind using the P2P logical network type for the OSPF.
"a no switchport to no switchport connection behaves more similarly to a P2P serial link than to an Ethernet. Maybe that is the rationale behind using the P2P logical network type for the OSPF."
makes sense...thanks for the help!