Re: [OSPF] Query regarding behavior of OSPF DR-Other's neighbor-State with BDR when DR fails, when DR down detection is delayed at DR-Other.

I agree with Tanmoy that this is proper behavior.
Thanks,
Acee
On Jun 18, 2013, at 10:00 AM, Tanmoy Kundu wrote:

Hi Bharath,
Few queries. As you mentioned "R2(The current BDR) detects the DR-down.", how did R2 sensed that R1 is down ?
    a. first possible option is dead timer expiry in R2. In that case R3 should also get the expiry soon and till that time the network won't converge. isn't that expected?
    b. Another option is having BFD session between R1 and R2, hence it comes to know. Why don't we run BFD between all the routers in the network ? As we know in OSPF the DR and BDR is not guaranteed.

This typical scenario is due to the DR-other is with priority zero. But when received hello packet from DR, both BDR and DR-Other should reset the dead timer. Even if we consider the link transmission delay and ASIC processing, the dead timer expiry difference between R2 and R3 should not be more than milisecond, isn't it ?

  *   Is  it right to trigger the NeighborChange event at R3? - I feel yes, other than few typical scenarios the network will at least not be used, until its converged. If Nbr-chg event is not sent then all DR other feels that DR is still active and the same will be used for forwarding the traffic. If someone in the network sensed that DR/BDR is down, why don't tell others immediately?
  *   Is this transition from FULL to 2-WAY  is expected? - As per RFC, DR others should not be FULL with other routers than DR and BDR, hence YES. It is expected.
  *   Can DR-Others flap adjacency with BDR if DR down detection  happens later than reception of new Hello from  the new DR? -
  *   Intuitively, it may seem desirable to continue to be adjacent to a neighbor as long as it is still DR or BDR. Is this a fair call? - As mentioned above, its not fair to use a disturbed or unsettled network for forwarding. Due to backlink check failure the LAN wont be used for forwarding during SPF. Hence as per me its proper behavior.

Thanks,
Tanmoy

On Tue, Jun 18, 2013 at 5:26 PM, Bharath R <rbharath@juniper.net<mailto:rbharath@juniper.net>> wrote:
Hi,
Please consider the scenario:

R1(DR)                                            R2(BDR)
|  I1                                               I2   |
|                                                            |
|______________________ |
                                |
                                |  I3
                                |
                      R3(DR-Other)

Here:
R1 is the DR.
R2 is the BDR
R3 is DR-Other, with DR-priority 0 and hence ineligible to become DR/BDR.

Following are the is the set of operations and a sequence of events:

  1.  Interface I1 is disabled. So, DR connectivity is lost to the rest of the network.
  2.  R2(The current BDR) detects the DR-down.

  1.  Declares itself the DR.
  2.  Declares BDR as NULL.
  3.  Sends out a Hello with DR as I2 interface address, and BDR as 0.0.0.0

  1.  R3(DR-Other) receives Hello from R2.

As per, section 9.2 of RFC 2328:
            o   One of the bidirectional neighbors is newly declaring
                itself as either Designated Router or Backup Designated
                Router.  This is detected through examination of that
                neighbor's Hello Packets.

            o   One of the bidirectional neighbors is no longer
                declaring itself as Designated Router, or is no longer
                declaring itself as Backup Designated Router.  This is
                again detected through examination of that neighbor's
                Hello Packets.

Triggers a NeighborChange event, which in-turn results in a DR election at R3.
Please note that at this time R3 has not yet detected R1 down.
Now result of DR-election at R3:
DR: R1
BDR: 0.0.0.0
Since R2 is no longer the BDR, R3 transitions from FULL to 2-WAY with R2(the new DR).
Of course, on detection of DR down at R3, R3 will elect R2 as DR and then again transition to ExStart, to Exchange to Full with R2.

Can you please let me know:

  *   Is  it right to trigger the NeighborChange event at R3 ?
  *   Is this transition from FULL to 2-WAY  is expected?
  *   Can DR-Others flap adjacency with BDR if DR down detection  happens later than reception of new Hello from  the new DR?
  *   Intuitively, it may seem desirable to continue to be adjacent to a neighbor as long as it is still DR or BDR. Is this a fair call?

Please correct me if I have missed out something.

Thanks and Regards,
Bharath R.

_______________________________________________
OSPF mailing list
OSPF@ietf.org<mailto:OSPF@ietf.org>
https://www.ietf.org/mailman/listinfo/ospf

_______________________________________________
OSPF mailing list
OSPF@ietf.org<mailto:OSPF@ietf.org>
https://www.ietf.org/mailman/listinfo/ospf