TUT220: Migrating NetWare Clusters to Linux Clusters

Another very interesting session. It is here that they explained WHY it is that Novell does not recommend running NetWare and Linux in the same cluster for very long. It is intended to be a migration step, not an extended state of being. This is very true in light of the various problems living in that environment poses.

Once you've added a Linux node to the cluster, you can't add a NetWare node. Or you can, but it isn't supported. The session demo had some troubles that could be related to this problem.

Extending NSS Pools requires reboots to get the visibility settled down. In pure NetWare, a simple "CLUSTER SCAN FOR NEW DEVICES" command will take care of it. In pure Linux, the method is less clear but is apparently well supported. But when mixed... reboots.

When adding trustees to a NetWare hosted volume, Linux will not see the added trustees until a manual process is kicked off to pick them up. This is probably one of the bigger problems. If the trustees won't migrate, that presents a big problem.

Services created in Linux will not migrate to NetWare. However, services created on NetWare will migrate to Linux. Usually. Some debugging of the startup scripts may be needed.

The folk in this session made a statement that broke my head. Apparently with the advent of eDirectory 8.7, Novell has started to recommend placing on all cluster-nodes replicas that contain any object that might need access to the cluster. One of the guys in the session asked:
I have a 12 node cluster. Are you saying Novell wants me to place a replica with tens of thousands of users in it, on each node? And that works?
To which the cluster guys said,
With eDirectory 8.7, it shouldn't have a problem with that. So, yes, that's what we're saying.
Right. So I went down to the eDirectory booth in the tech-lab after the session and asked their opinion of it. Apparently eDir IS robust enough to handle replicas with 30,000 user objects in them and do it on 12 nodes. My quibble is that my cluster nodes do very good yo-yo imitations, and I have a problem with replica holders doing that.

The clusters guys, and the eDir guys, both said that the cluster nodes really should have a replica on them that contains the cluster-context in eDir. That I don't have as much of a problem with, since that particular partition has only 1200-odd objects in it, and doesn't get updated all that often. We'll see if we do this once I get back into the office.

And in other news, Storage Routers are some cool things. In a few years when the price has dropped, we may be able to do the 'business continuity cluster' that TPTB want us to put together.

Tags: ,