Controlling DLSw peer selection

[jonathan.washington]

In the quest to provide durability for the network, we are using dual routers at each remote site. I am planning to use the ethernet redundancy feature to be able to use both routers when starting a circuit from the remote site, but the central site has two routers as well... and in addition to that I need to configure use of a completely separate disaster recover site as a backup to the central site.

With all that in mind I was wondering if it is possible to use costs on the peers at the central site to accomplish preferring one router over another and maybe even doing some load balancing, in conjuction with the primary / backup peer configuration?

Is it possible to say:

dlsw remote peer 0 tcp 1.1.1.1 cost 4 backup-peer 2.2.2.2 cost 3

dlsw remote peer 0 tcp 1.1.1.2 cost 2 backup-peer 2.2.2.1 cost 5

Must I choose between redundancy and load balancing?

[dixho]

What kinds of PUs are on the remote sites and central site?

The peer selection is determined by the DLSw peer which initiates DLSw circuits. Most implementation for PU4-PU2 scenario is that PU 2s initiate the LLC2/DLSw circuit. If the PU 2s are installed in the remote sites, the cost configured on dlsw remote peer statement on the central site routers has no effect on peer selection. It is because the central router does not initiate DLSw circuits. If this is a PU4 to PU4 connection, it does not work on multiple peer connections.

Most customers just do primary/backup peer without DLSw ethernet redundancy; OR customers just implement DLSw ethernet redundancy and connects the DLSw ethernet redundancy routers to multiple central routers. Of course, you can configure peer cost on the remote routers; so that the remote routers choose one of the central routers.

[jonathan.washington]

The remote sites are the initiators. I am wondering how best to load balance between central site A's router 1 and router 2, while being able to failover to disaster recovery site B's router 1 and router 2.

My guess was just to use your final suggestion and use peer costing on the remote site peers to direct traffic to A1 then A2 then B1 then B2.

I was hoping to get as granular as saying if A1 or A2 is not available then go chose B1 or B2. Can it be done?

[dixho]

I have a question on what you want to achieve.

Under normal situation, DLSw circuits are load balance between A1 and A2. If A1 goes down, do you want all circuit to go to A2? Or do you want DLSw circuits to go to B1 and A2? In other words, I do not know how to interpret "if A1 or A2 is not available, then go chose B1 or B2." It makes sense to me "if A1 AND A2 are not available, then go to B1 or B2."

If you want to load balance between A1 & A2 AND if A1 OR A2 is not available, then go to B1 or B2, you can set up 2 peers connection on each remote routers: A1 and A2. The cost of A1 and A2 on the remote routers are the same. B1 is set up as the backup peer for A1; and B2 is set up as the backup peer for A2. The cost for both B1 and B2 is same as that of A1 and A2.

However, I think that you may consider to increase the cost for B1 and B2; so that if A1 goes down, A2 takes all DLSw circuits. I think that it is a possbility that A1 goes down during non-disaster recovery time.

[jonathan.washington]

Sorry I was not very clear. I was not yet clear which way I wanted to handle the traffic and wanted to find out what the options were. I think your suggestion to have B1 and B2 costed higher is a good one, but after getting a little more information about the actual setup, the resource would only be found at site A or Site B not both, so equal costs would be fine. If there is no disadvantage, I won't worry about primary and backup peers, but just have multiple peers active and let DLSw decide which ones are in use based on where it finds the resource for which it is searching. ( But just for educational purposes... Can I actually configure both cost and a primary/ backup configuration on the same router? What would that command look like.)

I am still not completely clear on whether traffic to establish the circuit always flows from remote site to central site. I'm told that it could definitely flow in the opposite direction as well. That being the case, there is not a problem with configuring ethernet redundancy on both sides is there?

Thanks