Novell is now a part of Micro Focus

I have a NetWare 4 network and I want faster tree walking and name resolution.

Articles and Tips: qna

01 Oct 1998


Q.

Dear Ab-end: I have a NetWare 4 network and I want faster tree walking and name resolution. I have therefore placed replicas of the ROOT partition on all my servers. My assistant tells me this is not a good design. Is he right, and if so, what I can I do fix the situation?

—Arguing in Argentina

A.

Dear Arguing: Your assistant should be promoted. There are several reasons why yours is not a good design.

Reason 1. Partition operations (involving that partition) would require that every server be up and reachable. If one of those many servers were down or unreachable, the administrator could not split, join, or create a new partition, or merge a partition with its parent.

Reason 2. Each of these servers holds the same replica list and thus must be in communication with every other server. Time to synchronize will increase linearly as the number of servers increases.

Reason 3. Synchronization traffic will increase exponentially as more servers hold a copy of the root.

Reason 4. NDS cleanup processes occur when all synchronization has been successful. If a server is down or unreachable, cleanup processes will not run until all servers have been reached and synchronized to that time in the database.

Reason 5. Increased rights for the administrators to place a replica on each server will result in a decrease in security as more persons will need to have rights to place instances of the root replica on servers.

Reason 6. Every time a child partition of the root is updated (object and property changes, partitioning, and so on), every server that holds an instance of the child's replica list (for instance, every server holding a copy of root) will either have a Master, Read-Write, Read-Only, or Subordinate Reference of the child partition. Because the SyncUpTo time vector exists on the replica list, all servers holding the child replicas or subordinate references will be contacted for every change in that replica to update the time vector value in the replica list. Thus if all servers hold the parent (root), they all will be holding a replica for the child and they will all be contacted for changes to objects in the child partition. Remember that the subordinate replica contains a complete copy of the root-most container object for that partition. It holds all ACLs, attributes, partitioning information, time stamp values for the partition, pointers to all other instances of that partition, and so forth. It is much more than just a pointer.

It is important to remember that there are tradeoffs in placement of the root partition. Many instances may bring faster name resolution, but may sacrifice performance on the WAN due to synchronization traffic. This could also cause administrative problems by requiring that all servers be in communication and less efficient security. Though it is important to place instances of the root partition close to the users to facilitate efficient tree walking and name resolution, it is not prudent to place an instance of the root on all servers in the tree. Instances of the root partition should be placed in key locations, but not on every server in those locations.

* Originally published in Novell AppNotes


Disclaimer

The origin of this information may be internal or external to Novell. While Novell makes all reasonable efforts to verify this information, Novell does not make explicit or implied claims to its validity.

© Copyright Micro Focus or one of its affiliates