Re: Exit-Graceful Restart Condition

[Kishore Rao <[e-mail]>]

----- Original Message ----- 
From: "Padma Pillay-Esnault" <[e-mail]>
To: <[e-mail]>
Sent: Wednesday, May 25, 2005 12:46 PM
Subject: Re: Exit-Graceful Restart Condition


> ashok holla wrote:
>
> >Hi,
> >When myself and Sujay raised this issue, we thought about the
> >possibility which Pama and Acee suggested.
> >However, this will not work in all cases.
> >
> >Consider:
> >
> >TOPO------------------------ DUT---------------------------RTA
> >          Area0                                Area1
> >In this case, DUT will not get his pre-restart router of Area 1 as
> >there no other router in the area to give it to him.
> >Therefore, this solution fails.
> >In this case, suppose DUT had run the spf, he would have seen that the
> >type-3 summary lsa he had advertised into area 0, is no longer
> >consistent with the topology and then could have exited GR. However,
> >this consistensy check can be done only after grace period terminates
> >as RTA MAY come up within that time.
> >So, for this case, there seems to be no solution.
> >
> >
> You have not taken into account one major premise in thr rfc -
> For graceful restart to success you MUST have a helper to enable you to
> recover.
> In the case above you have
> 1. A double failure
> 2. No helper in the area
>
> With 2 - There can be no GR restart in the first place. Optimizing the
> detection is not possible if you do not even
>  have the list of your previous neighbors.

If no neighbors are detected after the interface Hello interval, wouldn't
exiting GR be advisable instead of waiting for grace timer to expire ?

Kishore

>
> >However, it does seem to be worthwhile to start deadTimers for all
> >nbrs listed in the pre-restart router lsa when we get it for the first
> >time.  (even though we have not received hello pkts from them). When
> >any inactivity timer fires on the restarting router, he can quit GR.
> >This can be done to impro convergence incase, topology is unstable.
> >
> >
> >
> I would put that as an option rather than a "must".
>
> Padma
>
> >Is our understanding correct?
> >
> >regards,
> >Ashok/Sujay
> >
> >
> >On 5/25/05, Acee Lindem <[e-mail]> wrote:
> >
> >
> >>Padma Pillay-Esnault wrote:
> >>
> >>
> >>
> >>>sujay wrote:
> >>>
> >>>
> >>>
> >>>>Hi All,
> >>>>
> >>>>Considering RFC3623, Graceful Restart for OSPF V2 I have a problem;
> >>>>
> >>>>Ex:
> >>>>
> >>>> (network)---- DUT ---- RTA
> >>>>
> >>>>(All in the same area)
> >>>>
> >>>>Assume DUT undergoes a Graceful-Restart, it sends out grace LSA's to
> >>>>RTA and other routers in the 'network'
> >>>>
> >>>>At the instant RTA recvs the Grace-LSA it also goes down.
> >>>>
> >>>>Is there a way in the RFC 3623 specified as to how DUT could Detect
> >>>>this topology change and Exit GR *before* grace period expiry??
> >>>>
> >>>>
> >>>>
> >>>>
> >>>I do not think the RFC covers this case. You are in a double failure
> >>>situation.
> >>>
> >>>I can think of a simple solution. When DUT reacquires its router-lsa.
> >>>It can determine its previous
> >>>neighbors. It can then decide that if within DeadRouterInterval that
> >>>neighbor did not  respond
> >>>to hellos to abort GR. This would be equivalent to the normal
> >>>situation when a router goes down and its
> >>>neigbors takes DeadInterval to bring down their adjacency
> >>>
> >>>
> >>I agree completely with Padma. One point, if RTA comes back up and
> >>originates a new router LSA
> >>then the DUT will detect the inconsistency (as specified in RFC 3623)
> >>with it's pre-restart LSA and
> >>terminate graceful restart.
> >>
> >>
> >>
> >>
> >>>Padma.
> >>>
> >>>
> >>>
> >>>>For other routers in the network RTA is still reachable and traffic
> >>>>Will be eventually dropped.
> >>>>
> >>>>Kindly correct me,
> >>>>
> >>>>Regds,
> >>>>Ashok/Sujay
> >>>>
> >>>>
> >>>>
> >>>>
> >>>>
> >
> >
> >
> >