DFS? It's In There Now that Microsoft has brought the Distributed File System (DFS) to NT 5.0, administrators can create a single export of their entire file system with a single share, with many servers participating in it as a child node of the directory structure. That is, once a client begins to browse a child node, all requests go to the server on which that part of the share resides, and the application won't notice anything different. DFS also can create child nodes that are replicated, providing fault tolerance as well as load-balancing. Any file created on one of the replicas is quickly duplicated on its twin.

Although we had some difficulty setting up the initial DFS structure, we managed to get it to work. In Beta 2, the directory initially must be empty, because the file replication service won't replicate previously existing files. But our replication directories weren't empty; we had created a replicated node in our DFS directory tree. We tried stopping and restarting the directory to make the service recognize that it had a new directory to replicate, but it didn't work.

Ultimately, we had to restart the two servers that stored the replicas to put the file replication service into gear. We frequently ran into such bugs during our NT 5.0 Beta 2 testing--the fixes should keep Microsoft busy right up until ship time.

To test DFS, we created a map from our DFS root to an NT 5.0 Workstation. Then we began to browse files in the DFS structure. By checking the properties on the child node directory, we determined which copy of the share the system was using. We unplugged the network from this server, and the client smoothly failed over to the other server. Even open files continued to operate as if nothing happened. Needless to say, we were impressed.

James E. Drews is a network administrator for the Computer Aided Engineering Center of the University of Wisconsin-Madison.

Send your comments on this article to him at jdrews@nwc.com or to Mike Lee at mlee@nwc.com.