Understanding eDirectory and Clustering
Posted: 13 Apr 2005
A reader recently asked for some insights on including replica servers in a
cluster:
"We are planning on setting up a cluster this summer. Here is our
environment: we have Netware 6.5 with four servers holding replicas. We also
have campus servers, where each site has its own server for file storage and
printing. Should we leave the replica servers "as is" or should we put the
replicas in the cluster?"
And here are the responses from several of our Forum experts:
The important thing to understand about clustering is that in order for
cluster failover to work efficiently, the nodes are intentionally less stable
than the cluster as a whole. This is an important concept to understand when
looking at cluster implementations, and how you design eDirectory around
this.
Let me explain. NetWare includes functionality that allows it to suspend a
running NLM that's crashed, without taking the server down. In a non-clustered
environment, this is a good thing, because the server stays up and can continue
to service requests for the non-failed services.
In a cluster environment, the node needs to go down hard. Otherwise, the
cluster heartbeat doesn't drop, and the services don't fail over.
eDirectory, by its very nature, is not a cluster-enabled service. It provides
fault tolerance through replication. (NLDAP, on the other hand, is a clusterable
service, but you must fail the IP addresses over).
Cluster Container Object
There is a cluster container object that is used for the purposes of managing
the cluster itself. This object needs to be available for every node in the
cluster in order for certain things to work properly. Among other things, the
membership of nodes in the cluster is tracked using something called a "panning
ID," which is stored in this cluster container object.
As I understand it, the best practice is to replicate the cluster container
(as a partition root) to each node in the cluster, or at least to a number of
nodes in the cluster. Other partitions would be replicated as per usual,
ensuring there is sufficient replication from the standpoint of the directory to
provide connectivity/authentication should any particular node in the cluster go
down.
In other words, apart from the cluster container object's requirements,
eDirectory replication itself doesn't really need to be treated any differently
in a clustered vs. non-clustered environment, because the service itself is
inherently not cluster-aware.
Other Tips
1. You can treat cluster servers just like any other server as far as replica
placement goes. However, make sure you have enough replicas to take in to
consideration that an individual cluster server might go down more frequently
than a non-clustered server.
2. When we upgraded from a group of 5.1 servers to a 6.x cluster, we just
spread the replicas out across the cluster servers and it has all been working
fine. Also, for more help setting up your cluster, be sure to check the
cluster-services Novell Support Forum.
3. The cluster container itself has to be treated somewhat specially, but not
because of eDirectory requirements - it's needed for the cluster's own
operation.
4. There is a way to "cluster" clusters for multi-site redundancy. It's not
cheap to do (you need to have multiple SANs), and because of the complexity, it
may require a consulting services engagement from Novell.
|