Hello,
as per RFC2328 that defines OSPFv2 for IPv4 during the exstart state in the OSPF per neighbor state machine (finite state machine) a Master is elected and the other device is a Slave.
This is just to decide the initial sequence number. The highest router-id wins the master/slave so in normal conditions is the current DR the Master.
OSPFv2 is a routing protocol for IPv4 but its own encapsulation is IP protocol 89.
Once elected the master sends LSA headers grouped in packets to the Slave
The Slave can acknowlege correct reception of previous LSA headers.
So the goal is to to synchronize the databases between the DR and the possibly new router.
Emulating what TCP does we could say:
>>OSPF uses a window with size one packet the packet is sent at MTU with DF bit set and TTL=1
the slave can Ack each LSA update sent by the master by sending the seq-number of last successfully received LSA update.
But also the slave can have something new to tell to the DR.
Both routers end the loading phase with a selected list of LSA headers that is a subset of received LSAs and use LSA request to ask for the whole content of each new or more update LSAs it learned in the exstart state.
When database are synchronized because each router has received the missing info the two databases of R1 and R2 are in full state and R1 and R2 are considerered to be fully adjacent.
In LAN with multiple routers to save on all this activity a Designated and a Backup Designated routers are elected and listern to 224.0.0..6 All OSPF DR in local subnet.
All other devices listen to 224.0.0.5 All OSPF routers in local subnets.
To be noted DRs and BDRs listen to 224.0.0.5 too.
In modern networks we use Ethernet variants for example an LACP bundle of 4 x 10 GE member links
To be noted in old networks with 50 devices in the same LAN 20 years ago the DR was the device with the highest uptime that could be a Sun Solaris or HP workstation with OSPF daemon active.
the command
ip ospf priority in interface mode
can be used to avoid to have router to be elected DR in many LAN segments.
All this happens within a single OSPF area.
OSPF is link state within a single area.
Communication between areas require the use of the backbone area 0.0.0.0 or simply 0 all devices acting as bridges between area 0 and other areas are called ABR = Area Border Router.
All user traffic and all routing information from area X to area Y have to go via the respective ABRs routers
so we can say
R4 ------ R2 -------- area 0.0.0.0.0 R1 ---------------- R3 ---R5
area1 area 0.0.0.0 area 99
so R4 is internal to Area 1
R2 is ABR (0,1)
R1 is a backbone router internal to area 0
R3 is ABR (0,99)
R5 is internal to area 99.
ABRs mantain a different LS DB for each area they have a leg into.
In my humble opinion the usage of areas is highly recommended regardless of the number of routes.
The reasons are the following:
- control of internal routes OSPF iintra area OSPF inter - area can only happen at area border
- external routes should be minimized as they cannot have fine control as they propagate everywhere.
_ NSSA area types are to be preferred over stubby areas
- BUT NSSA require the ABR(s) to be configured to inject a default route into the NSSA area, stubby areas ABR nodes should inject a default route into the stubby area to cover the fact that all external routes LSA type 5 are blocked
it is better to use passsive-interface + other commands instead of redistrubuted connected as the second method produces external LSA type 5 routes
Hope to help
Giuseppe Larosa
