Flaky WAN, but EIGRP stays up

[m.glosson]

Sometimes our MPLS connections stay up, but drop many packets. We have to manually switch over to the backup DMVPN because the EIGRP neighbor relationship never actually goes down, but performace is terrible. So we manually switch to the backup, call the SP and wait for them to fix their issue before switching back.

What is something proactive within the router that can catch this type of situation and react before the users inform us?

If someone has a nice ip sla or event manager config for this that they would be interested in sharing, that'd be cool. Or some other method... whatever.

Thanks

[Philip D'Ath]

You could consider using Pfr (performance routing).  This lets you control routing based on lots of factors like packet loss, latency, jitter, etc.

http://www.cisco.com/c/en/us/td/docs/ios-xml/ios/pfr/configuration/15-mt/pfr-15-mt-book/pfr-eigrp-mgre.html

Any chance you have an access much bigger than the bandwidth you have?  E,g.  a 100Mb/s access and 20Mb/s of bandwidth?  In which case you need to look at QoS.

[m.glosson]

Thanks for pointing me at PfR. I love how it thinks IOS is so awesome: "Cisco Performance Routing takes advantage of the vast intelligence embedded in Cisco IOS Software..." No points for humility there!

It's not a bandwidth issue--it's that some links flake out from time to time with the MPLS provider--they are not hard down, but they drop 40-50% of the packets... good enough to not let the EIGRP hold timer expire, but not good enough to have happy and efficient TCP sessions. But it's not a bandwidth issue (for sure).

Regardless, I'm looking at solutions, and this PfR seems like it would be one of them, if I have the courage to bite into such a big meatball.

[Philip D'Ath · Accepted solution]

Don't look at anything else.  Pfr is what you should be using.  Use a packet loss trigger.

[Simon Brooks]

How do you manually switch the traffic now?

[aleksandarsofranic]

How about IP SLA?

So what i was thinking is to use ip sla towards the eigrp neighbor, and tie it down to a default.

So as soon as you loss 2-3 icmp packets it should fail over to your backup, also use this

track 1 ip sla 1 reachability
..delay down 90 up 90

This will make sure that the delay before relaying traffic to you primary link is 90 seconds, so you don't keep going up and down if it is a congestion problem

Hope it helps

cheers!