BFD with LDP

[Sherif Atef Ahmed Ismail]

Dears

Would like your assistace please regarding below issue

I am having directly connected primary/backup PE and each is connected to a seperate P router. On the uplink going to P router, we are configuring BFD and LDP-IGP sync delay so when interface flaps routes will be received with maximum metric till the delay period expire.

During this period, directly connected backup PE should be used

All this is working fine .. however problem if OSPF flapped due to BFD, LDP won't sense this failure as LDP is not registered with BFD

As a result this delay wont be applied (this delay is useful so to make sure that link is somehow stable before using it again)

This can be solved with BFD dampening however found feature only available with ASR1K and I am having 7600 routers

So question .. is LDP with BFD supported and If yes how to configure it

If you have other ideas, please share

Many Thanks

Regards

Sherif Ismail

[Harold Ritter]

Hi Sheriff,

Can you tell us what is the specific issue. When bfd detects the link failure, ospf is notified and it reconverges around the failure. ldp is not notified but it does not matter at this point as traffic as already been shifted to a new path. When the link comes back up, ospf reestablishes the neighbor relationship to the P router and advertises the link with a max metric until ldp session has been reestablished and labels have been exchanged.

You might want to look at ldp session protection to minimize the time you wait after ldp. With ldp session protection, the session would not go down as long as there is an alternate path between the two ldp peers.

Regards

[Sherif Atef Ahmed Ismail]

Hi Harold

First many thanks for your time to reply/assist

Below is the topology in concern

Each PEs has 2 MP-iBGP session to both RR going from its uplink.

Uplinks are Ethernet interfaces so they rarely go down

When primary uplink go down, traffic should go to backup uplink via interconnection between both routers

My only issue is with short successive flaps (1 sec<). During these flaps LDP session is still up cause as I understand it will only go down when its timers expires so LDP-IGP delay is never applied. This means when OSPF is up again link will be immediately used without applying any delay.

Yes traffic should not be lost cause during these flaps traffic should go to backup router thanks to OSPF IP-FRR fast convergence however I wish to give some penalty to flapping uplink so to make sure link wont be used again till somehow it is stable

Believe interface dampening won't help here as interface never goes down. Also to me LDP IGP-delay is better cause If backup path is down during this delay time, link on primary router can be used cause there will be no other better metric received

However problem now that BFD does not notifiy LDP when it goes down. So short flaps will not be sensed by LDP

Many Thanks

Regards
Sherif Ismail

[Vivek Ruhil]

Hi

to avoid a flap situation with ldp-igp sync, you could use the holddown time.

mpls ldp igp sync holddown . I would suggest to use 5 for a start as the timer value.

LDP is not supported as a client of BFD, its the routing protocols that become clients.

So, your OSPF will be notified and with this timer in place you should not see any flaps.

[Sherif Atef Ahmed Ismail]

Hi Vivek

Thanks for your reply

Actually I wish LDP would also flap cause only then LDP-IGP delay would be applied .. To me this delay would acheive a similar effect to that of dampening

Here is an example

R1#sh bfd neighbors details | i Uptime

Uptime: 02:12:51

R1#

R1#sh ip ospf neighbor detail | i up

    Neighbor is up for 02:12:56

R1#

R1#sh mpls ldp neighbor | i Up

         Up time: 1d15h  <<-----   didnt sense the short flap

R1#

Maybe I am moving in a wrong direction ... In above scenario how could I acheive a dampening effect .. i.e when primary links flap (BFD) and traffic moves to backup link .. traffic wont go back again to primary link unless a penalty/delay is applied to it

Regards

Sherif Ismail             

[Vivek Ruhil]

Hi Sherif

There is no concept of a penalty/delay with BFD. However, you could use the timer value from my above post to minimize the impact of such flaps.

Say:

1. At T1 there is a flap on primary link and your traffic shifts to secondary link.

2. Now, the primary link comes up at T2.

3. You have configured the timer as 5 secs and hence the router will wait for 5 secs before advertising the primary path in IGP and also allowing LDP to use that path.

4. Supposing T2 was just a flap and the interface has again gone down.

5. Again the IGP will be advertised with full metric making the path non-usable.

6. In short, you could save your traffic from a flapping link by using a holddown timer.

Alternatively, you could use IP dampening on the interface which will give it a penalty on every successive flap and keep it down. But, this will keep the interface completely down rather that the LDP-IGP Sync funda.

Regards

[Sherif Atef Ahmed Ismail]

Hi Vivek

Many thanks for your follow up on the thread

As I understand "mpls ldp igp sync holddown" defines the time IGP would wait for LDP to be up cause by default if LDP is not up OSPF will wait forever and will never be up. So using this timer will make IGP be up after timer expiers even if LDP is not up ... hope my understanding is correct

In my case which is the very short flaps intervlas < 1sec , LDP is never down so actually this command according to my understanding will not help

Also as I know "mpls ldp igp sync delay" will never be applied to IGP unless LDP is down

So when there is short loss of connectivity determined by BFD, OSPF is notified as it is BFD client but LDP not

Once OSPF is up again it will advertize metrics again with normal metric without applying any delay .. Thus primary link would be used immeditely

If interface went down during these flaps then all would be perfect cause in this case LDP would be down and IGP would advertize/receive prefixes with max metrics during the delay time .. during this time backup link would be used which is the desired design

I did not test it but I believe Interface dampening also will not help cause if interface didnt go down which is usually the case with Ethernet WAN interface, dampening would not be applied

If I am not mistaken, I believe the only solution now would be BFD dampening but according to Cisco naviagtor feature it is only supported on ASR1K .. however with this solution will be in risk of total isolation if backup link went also down during the dampening perid of primary link

What do you think ?

Many Thanks
Regards

Sherif Ismail

[Vivek Ruhil]

Hi sherif

Apologies for the late reply.

"

As I understand "mpls ldp igp sync holddown" defines the time IGP would wait for LDP to be up cause by default if LDP is not up OSPF will wait forever and will never be up. So using this timer will make IGP be up after timer expiers even if LDP is not up

"

I'm afraid that ldp-igp sync doesn't work the way you described above. Following Steps occur:

1. LDP and IGP goes down.

2. somehow, IGP is up but LDP is not.

3. Now IGP(OSPF) will start advertising its link cost as the maximum value rendering this route to be unusable. PLEASE NOTE IGP SESSION WILL BE UP.

4. Default behavior is that as soon as LDP comes back up, IGP immeidately withdraws advertising maximum metric and starts advertising he correct metric. This situation can cause downtime of there are flaps in the network.

5. The holddown timer is used to ensure that IGP waits for LDP to be up for the specified interval before it stops advertizing the maximum metric and makes the route usable.