Optimising WAN link detection period

[jedellerby]

Hi,

I'm trying to work out how I can optimise WAN link detection failure for a frame relay circuit.

I have a site with two routers each with a PVC back to a central site. The routers are currently setup with HSRP on the local LAN with interface tracking enabled on the frame relay serial link. The HSRP timers are set to 1 3. Static routing is configured on both routers.

The aim is to try and avoid any outage to the central site greater than 2 seconds. The HSRP timers could be tuned downwards to help achieve this.

However, frame relay LMI keepalive (or Cisco end-to-end keepalive) monitoring doesn't have low enough timers to acheive this reliably.

I don't beleive I could get low enough figures by running a routing protocol as the timers would still be too high. I've looked at IP SLA and once again the best frequency is 1 second for ICMP monitoring hence too slow. I also checked out Cisco Optimized Edge routing but I don't think that's applicable. I also looked at Bidirectional Forwarding Detection, and this isn't available on the lower platforms.

I think I need some monitoring protocol that will give me end to end feedback with msec timers associated. I also need to be able to feed this information back into HSRP or the routing engine (static or dynamic).

Does anybody know if this is feasible and how it can be acheived?

Thanks

Jed

[tdrais]

Never actually configured this in anything other than a lab but you in theory could see a outage in 1 second using ospf fast hellos

http://www.cisco.com/en/US/products/sw/iosswrel/ps1839/products_feature_guide09186a00801039b1.html

[sundar.palaniappan]

Jed,

IMO, a routing protocol like EIGRP will probably give you the quickest covergence time possible. You can set the hello/hold interval to 1 second and non-receipt of single hello will cause the neighbor to go down via serial link and that I hope that is acceptable to you.

R4(config-if)#ip hold-time eigrp ?

<1-65535> Autonomous system number

R4(config-if)#ip hold-time eigrp 1 ?

<1-65535> Seconds before neighbor is considered down

R4(config-if)#ip hold-time eigrp 1 1 ?

R4(config-if)#ip hold-time eigrp 1 1

HTH

Sundar

[jedellerby]

Thanks for the first response regarding OSPF fast hellos. I see I can achieve this with both IS-IS and OSPF which is good. However, my first preference would be to go for EIGRP as this is the legacy routing protocol running in various other parts of the network. Introducing OSPF brings new management headaches which I will try to avoid. However, I think OSPF could achieve what is required.

Sundar ... thanks for the feedback but I think the EIGRP idea of using a hold interval of 1 second is too risky. This has the potential to trigger more instability over the WAN links as only a single packet need be lost to cause a topology change. Most routing protocols are designed to avoid this type of issue by working on multiples of the hello timers. Thanks for the idea but I don't feel I could recommend this approach.

My perfect scenario is to come up with something that is dynamic routing based as this particular part of the network is static routing based. I think it maybe wishful thinking however !

Any more ideas?

Jed

[jedellerby]

Does anybody know if the following appraoch might work:

- Run UDP jitter monitoring continuously with an 'interval' of say 250ms and a timeout of 500ms

- Setup a threshold monitor for lost packets and trigger after 'n' lost packets.

- Feed the trigger into HSRP to failover

What I'm not sure about is whether SAA/IP SLA can really do this. I think the jitter bit is ok but not so sure how the triggering works, possibly this can be achieved by reaction-configuration.

I suspect the the 'track' command could be used to determine if the SAA process becomes 'active' which can be set to happen based on the reaction-configuration.

I can see in theory I might be able to make this work, but I suspect there will be some issues in the low level detail!

Thoughts anybody ?

Jed