Re: [Ospf-wireless-design] Another example for self-selected MPRs

["Madhavi W. Chandra" <[e-mail]>]

On Tue, Mar 29, 2005 at 06:14:16PM -0800, Richard Ogier wrote:
> 
> To help understand how my proposed solution uses Hellos to learn
> 2-hop neighbor information and to update self-selected MPRs (or MDRs),
> I would like to present another example.
> 
> Let's assume that (differential) Hellos must be used to learn
> 2-hop neighbor information, e.g., so that partial-topology
> LSAs can be generated.  Also, assume that Hellos are sent
> every 2 seconds, and that no early Hellos are sent.
> 
> First, I want to point out that, for self-selected MPRs/MDRs, only
> DRs and BDRs need to advertise their full neighborhood (periodically)
> in Hellos. This is an additional benefit of using self-selected MPRs,
> because they are in the "middle of the action", i.e., they can directly
> see both the upstream neighbors and the downstream neighbors.
> However, for this example, I will assume that all nodes periodically
> advertise their full neighborhood in Hellos.
> 
>       1
>      / \
>     /   \
>    2 --- 3
>     \   /
>      \ /
>       4
> 
> 
> Assume that initially node 3 is an MPR, either self-selected, or
> neighbor-selected by node 1.
> 
> Now assume that node 4 detects the failure of its link to node 3 at
> time 0, either from a link-layer failure notification, or from
> not hearing Hellos from node 3 for RouterDeadInterval.
> 
> Then node 4 will send a Hello at time 2 (worst case) that indicates
> node 3 is no longer a neighbor. Assume that node 3 can still hear
> Hellos from node 4, so both nodes 2 and 3 will receive the Hello.

Actually, the backup Overlapping Relays (node 2) will go into action 
since it wouldn't have received node 4's ACK.  This will happen at 
(Pushbacktimer + jitter), which is latest by 
(RxmtInterval/2 - propagatation delays).  Even taking
the default RxmtInterval of 5 seconds, puts this time on par, if not
sooner, with the 2second Hello time example.  So, the rest of the example 
of Node 2 knowing it is an Active Overlapping Relay at time 6 is moot.  
Jitter is used for backups to avoid multiple backups firing at the same 
time.  

Please keep in mind that the backup mechanism in Overlapping Relays is 
a powerful yet simple feature and integral to the operation, and should 
not be overlooked.

Thanks,
Madhavi
 
> If MPRs are self-selected, then node 2 will select itself as an MPR
> at time 2, when it receives the Hello from node 4.
> 
> If MPRs are neighbor-selected, then node 3 will send a Hello at
> time 4 (worst case) indicating that node 4 is no longer a neighbor.
> When node 1 receives this Hello, it will select node 2 as an MPR,
> and will send a Hello at time 6 including the updated MPR.
> Node 2 will then know it is an MPR at time 6, compared to time 2
> for self-selected MPRs.
> 
> (I know that non-active overlapping relays will forward MPRs
> during this recovery period, but with an added delay of
> PushbackInterval, and possibly additional overhead if
> mutiple non-active relays flood the LSA at nearly the same time,
> a real possibility if the number of non-active relays is large.)
> 
> Richard
> 
> 
> 
> 
> _______________________________________________
> Ospf-wireless-design mailing list
> [e-mail]
> https://www1.ietf.org/mailman/listinfo/ospf-wireless-design

_______________________________________________
Ospf-wireless-design mailing list
[e-mail]
https://www1.ietf.org/mailman/listinfo/ospf-wireless-design