ATM pvc failover issue

[bsparker418]

Ok I have two routers in to different cities both talk to each other over AT

M w/ BGP I have a pvc to 1 client at both locations, They don't want me to l

oad balance between the two pvc but want a failover senario. So city 1 goes

down all traffic goes to city 2, I know I have to weight this somehow but I

am not sure where. both routers know about the customers network as a local

link so we tried weighting the primary link to make it go there first. that

did not do the trick, here are the general config

router in city 1 (main)

router bgp MY AS

no synchronization

bgp default local-preference 500

bgp log-neighbor-changes

network MY NETWORK

network CUST NETWORK

redistribute connected

redistribute static

neighbor CUST LINK1 remote-as CUST AS

neighbor CUST LINK1 weight 200

no auto-summary

router in city 2 (FAILOVER)

router bgp MY AS

no synchronization

bgp log-neighbor-changes

network MY NETWORK

network CUST NETWORK

redistribute connected

redistribute static

neighbor CUST LINK2 remote-as CUST AS

no auto-summary

the link on city 2 is shutdown to make sure no traffic will go over that rig

ht now

I am thinking I have to weight the ip route statement but all I get is a met

ric and I don't think that will do it. Need a clue to point me in the right

direction

Thanks in Advance

[vmoopeung]

I don't see any problem in the configuration, You can use local preference statement to do the same. weight is not actually needed. please refer the document below.

http://www.cisco.com/univercd/cc/td/doc/cisintwk/ics/icsbgp4.htm#4957

[bsparker418]

just as a update I have updated site 2 with a local pref lower than site 1 but I still have traffic moving over site 2 when I enable the pvc. this is bad thing in my situation becuase of the app on the customer side freaks out when more than 2 paths are provided ( not my app ) and it gets timeouts, I have recieved another hint with route maps here is the config i am thinking about

Interface fast/ethernet 0/1

ip policy route-map CUST-FAILOVER

Access list 1 permit ip 1.1.0.0 0.0.255.255

Route-map CUST-FAILOVER permit 10

match ip address 1

set ip precedence priority

set ip next hop CITY 1

Route-map CUST-FAILOVER permit 20

match ip address 1

set ip precedence critical

set ip next hop CITY 2

[Richard Burts]

The biggest problem with this proposed solution is that both instances of the route map refer to the same access list. Any packet that matches in CUST-FAILOVER permit 10 will never get to CUST-FAILOVER permit 20 because once there is a match the route map does not look any further. And any traffic that did not match CUST-FAILOVER permit 10 will also not match CUST-FAILOVER permit 20.

More fundamentally I think you need to consider carefully whether you want to try to solve your problem with policy routing (the current suggested solution) or in the routing protocol.

I would suggest that the best solution is in the routing protocol. One of the weaknesses of the policy routing solution is that it only affects how you send traffic out and can do nothing for how traffic is sent to you.

Another potential weakness of the policy routing solution is that you are attempting to forward to some remote non-customer network. If you policy route you might very possibly have a solution where site 2 has a packet and policy routing send it to site 1. But the routing table at site 1 might have the best route indicated as through site 2, so site 1 sends to site 2 who sends to site 1 and it goes around till TTL expires (a particularly interesting routing loop).

If you really want failover and no load sharing I suggest that you need to use BGP to do that. I also suggest that you not only try to control how you send traffic out to your neighbors but that you try to control how your neighbors will send traffic to you. Otherwise you may have a situation where you send all traffic out through site 1 but much of the return traffic comes through site 2.

[bsparker418]

Ok I might be miss reading your Statement but I already am running BGP between my two routers. I have set BGP to have a local pref at each site city 1 being higher that city 2, But even after that when I enable the link at city 2 I get traffic moving over that link which in turn breaks the app that we use ( the only thing we use this link for is this app no other traffic goes on it ). From What I have been readin route-map is the only way I can force traffic going to this cust site using city 1 link then if it goes down use city two. thus i came up with this config

Interface fast/ethernet 0/1

ip policy route-map CUST-FAILOVER

Access list 1 permit ip 10.10.0.0 0.0.255.255

Route-map CUST-FAILOVER permit 10

match ip address 1

set ip precedence priority

set ip next hop 10.250.14.2

Route-map CUST-FAILOVER permit 20

match ip address 1

set ip precedence critical

set ip next hop 10.250.13.2

If you see a better way to achive this please let me know, I just would like to know if this config will work they way I want it before I but it in production and the break everything :)

Thanks in advance

[Richard Burts]

First of all, this config will NOT work. As I said in my previous post "Route-map CUST-FAILOVER permit 10" and "Route-map CUST-FAILOVER permit 20" are redundant. They both have the exact same match clause "match ip address 1" and that is a big problem. No packet will ever be affected by sequence 20.

I interpreted your post that the policy routing solution was an alternative to the BGP solution. I now understand that it was intended to supplement the BGP solution. Even if you get it to work I am not sure that it will solve your problem.

I have been thinking more about the problem you describe where the application has problems when there is load sharing. It makes me wonder if the problem is about load sharing how you send data out or if the problem is about load sharing how data is sent to you.

Both of the solutions you have talked about are about how you send data 0ut. I have a problem understanding how the application could have a problem in load sharing outbound because I do not see how the application would know how data is sent out. The application is much more likely to recognize load sharing how data is sent In. But neither of the solutions you have talked about deal with how data comes back. To control how your neighbors send data to you, it will be necessary to consider solutions like MED, or community, or AS prepending.

Perhaps you could tell us a little more about the application and what kind of problem it has when there is load sharing (and how does it know).