asymmetric routing

dvanzee · ‎08-29-2010

I have two sites with MPLS connections (SiteA) and (SiteB) these sites are also connected via a man connection to each other. In addition, I have 70 remote sites connecting into MPLS. If one of the remote sites goes down (circuit failure). SiteA and SiteB take over the network range of the remote site. When the remote site comes back up, SiteA and SiteB hold onto the network address causing routing issues for the remote site. The only way I have been able to fix this is by placing a route map on site A and site B mpls router denying the remote sites network range "in' and clearing BGP. During the period site A/B take over the address, the network range in question show the network range being learned via EIGRP until site a MPLS router where learns it from BGP from site b. Below is the connection and routing for siteA/B.

BGP AS 64551 BGP AS 64441

SiteA MPLS RTR SiteB MPLS RTR

EIGRP 100 EIGRP 100

| |

|___________L3MAN_______________|

site a
router eigrp 100
redistribute bgp 64551 route-map BGP-TO-EIGRP
network xx.xx.xx.33 0.0.0.0
network xx.xx.xx.128 0.0.0.7
network xx.xx.xx.1 0.0.0.0
default-metric 1000 100 255 1 1500
distribute-list 10 out GigabitEthernet0/0/0
no auto-summary
!
router bgp 64551
no synchronization
bgp log-neighbor-changes
network 0.0.0.0
network xx.xx.146.0
redistribute connected
redistribute eigrp 100 route-map EIGRP-TO-BGP
neighbor xx.xx.xx.25 remote-as 65000
neighbor xx.xx.xx.25 route-map sitea-BGP-ROUTES-IN in
neighbor xx.xx.xx.25 route-map siteb-ROUTES-OUT out
default-information originate
distribute-list 11 in GigabitEthernet0/0
no auto-summary

site b
router eigrp 100
redistribute bgp 64550 route-map BGP-TO-EIGRP
network xx.xx.xx.24 0.0.0.7
network xx.xx.xx.32 0.0.0.7
network xx.xx.xx.38 0.0.0.0
default-metric 1000 100 255 1 1500
no auto-summary
!
router bgp 64550
no synchronization
bgp log-neighbor-changes
network xx.xx.xx.32 mask 255.255.255.240
network xx.xx.xx.38 mask 255.255.255.255
redistribute eigrp 100
neighbor xx.xx.xx.85 remote-as 65000
neighbor xx.xx.xx.85 route-map siteb-BGP-ROUTES-IN in
neighbor xx.xx.xx.85 route-map sitea-ROUTES-OUT out
default-information originate
no auto-summary

I suspect when the remote site fails, site b advertises the remote network into BGP, thus site a learns it from BGP and the EIGRP timer can never expire that is why putting the route map "in" on both routers clears the routes out of the routing table. Can someone tell me what I have setup incorrectly and how to fix it?

gatlin007 · ‎08-29-2010

In the BGP-to-EIGRP route-map you need to set a 'tag value' on the route. In the EIGRP-TO-BGP route-map you need to deny routes with the tag previously applied. This will prevent the route 'latching' behavior you are seeing.

Something like the following:

route-map BGP2EIGRP permit 10
set tag 86
exit

route-map EIGRP2BGP deny 10
match tag 86
exit

route-map EIGRP2BGP permit 20
exit

Chris

dvanzee · ‎08-31-2010

Chris - Thank you for the reply. After researching this issue further, the symptoms have changed. Here is what I see.

Normal operation - Site A and Site B learn the 70 remote office networks from their respective MPLS routers.

Failure - If one of the remote sites goes down, once the network comes back up the route/network is learned and preferred from site a MPLS. Meaning if you go onto site B MPLS router the remote network is in the BGP routing table, but is not the preferred route "*" not "*>" , however the routing table shows the path to this remote site going over the L3 man link and out site a MPLS router. Again this only happens when the office goes down, to clear this issue I have been using route maps to deny this network "in" to both sites and soft resetting BGP seems to clear the issue. Thoughts?