Before discussing these strategies for replacing NIS with LDAP, it's worth understanding something about the beast we're trying to replace. NIS is most commonly used to distribute system password and account maps (i.e., /etc/passwd and /etc/shadow) to client machines. It's also used to distribute the information from many other system files, such as /etc/hosts, /etc/services, /etc/group, and /etc/networks. It can also distribute a number of files that control the automatic mounting of remote file systems; and with the appropriate wizardry in sed and awk or Perl, it can be coerced into distributing almost any kind of data that can be represented in a text file.
 This discussion is necessarily very brief. If you need more information about NIS, see Managing NFS and NIS, by Hal Stern, Mike Eisler, and Ricardo Labiaga (O'Reilly).
In the NIS world, the master copy of any shared data resides on a master server, which distributes the data to slave servers. Clients, which are organized into NIS domains (not to be confused with DNS domains), can then access this information from any NIS server, master or slave, that services their domain. The NIS master acts as a directory system agent (DSA) that provides information to clients, which use this information to perform tasks such as authenticating users (i.e., the passwd map) and locating other hosts on the network (i.e., the hosts map).
The NIS information model is also characterized by a flat namespace. To use the passwd.byname map as an example (this map represents the /etc/passwd file, indexed by username), there can be only one login name of jerry. To work around this deficiency, NIS groups client machines into NIS domains, each with its own set of maps (i.e., its own set of virtual administrative files). So, two users with the login name jerry can coexist if they can be placed in different domains; for all practical purposes, different NIS domains are different directories (even though they may be served by the same server).
In contrast, LDAP allows you to create a hierarchical namespace to manage these users. Let's assume that we'll use the RDN of a node as its login name. LDAP can then maintain multiple users with the same login name if we make sure that each user belongs to a different parent node (see Figure 6-2).
One basic rule of system administration is that users should not notice any loss of service when you implement changes. A user does not need to be aware of where their account information is stored. It makes little difference to them, as long as they can access necessary network services. If a change results in a downgrade of service for users (no matter how big of a win for the system administrators), you'll almost certainly be forced to rip it out and go back to the old system; eventually, you'll get tired of answering all the help desk calls. Fortunately, the flexibility of the PAM and NSS interfaces can do a lot to insulate users from a change in information location.