Load Balance & Fault Tolerance

[jrmoraes77]

I need do design a solution for load balance the DLSw traffic between 4 central routers and, if this 4 routers fail (oe wan fail) all peers and circuits need to be restablished on other site with other 4 routers.

To balance the traffic I will use the DLSw circuit count. To provide fault tolerance between sites I thinking to use backup peer.

My question is, "circuit count" will work togheter with "backup peer" ?

Thank´s in advance.

[dixho]

I would like to make sure that I understand the topology. The routers in remote sites have 4 DLSw peers defined. They also have 4 backup peers defined.

If this is the case, DLSw works.

However, I would like to raise some questions. What happens if only one of the DLSw peer disconnect. (i.e. 3 DLSw peer connections stay up. Another DLSw peer connection goes to a backup site) This may create a problem for you. How often do you sync the data the mainframe in the primary site and backup site? If the data in the mainframe are not in sync while you have an outage, end users will access inconsistent data.

In another scenario, all 4 peer connections disconnect. All end users will access data from the backup host. Is there an issue there?

[jrmoraes77]

Your understanding about this environment is correct. Your comment about are relevant too. To workaround this issue about backup peers I can define 7 backup peers in each remote router, look this:

Central site 1 : routers A, B, C and D

Central site 2 (backup): routers E, F, G and H

Remote router

- 4 peers established to 4 central routers (A, B, C and D) with circuit count

- backup peer to B, C, D, E, F, G and H

In this case, if if router A fail, the remote router will establish a new connection to router B, C, D, E, F, G and H in this sequence.

All remote routers need to be well configured to this work fine.

What you think about ?

[dixho]

Only one backup dlsw peer is allowed. I cut and paste the following when I try to define more than one backup peer:

c3-2500(config)#dlsw remote-peer 0 tcp 2.2.2.2

c3-2500(config)#dlsw remote-peer 0 tcp 3.3.3.3 backup-peer 2.2.2.2

c3-2500(config)#dlsw remote-peer 0 tcp 4.4.4.4 backup-peer 2.2.2.2

%Primary peer already has backup defined

There are a number of approaches:

1. Remote routers have 8 peer connections. The cost for A, B, C, and D are lower than that of E, F, G, and H. Normally, the circuits are distributed among A, B, C, and D. Even one or more than one of A, B, C, and D goes down, the rest will take the load. If all A, B, C, and D goes down, E, F, G, and H will take all the circuits.

2. Slightly different than 1. Instead of making E, F, G, and H are permanent DLSw peer connection, make E is a backup peer for A, F is a backup peer for B, and so on.

3. Just another idea. Have you considered SNASw using HPR/IP? It may take you a while to set up on the host. However, this is the way to go because IBM has stopped selling 3746/3745. All SNI link will eventually go to HPR/IP.

[jrmoraes77]

I understand your alternatives and will test the #2 in a lab environment. The #1 I´m not sure because I don´t know how the WAN environment will be made.

The alternative #3 is the future of this network and will be deployed in next year, by now I need to find some solution based on backup peers or something like this.

Thank you for your help (again)

Best regards