On one extreme of the network host management scale is your basic MS-DOS PC, with virtually no knowledge of the network. Windows, and its newer derivatives in particular, recognize the network as an integral part of the software platform, and permit centralized control of users, directory permissions, and the like. A step beyond this is a typical UNIX-based X-Windows workstation, which provides similar permission-based access control, and allows a program running on a remote system to display a window on the screen and manipulate its contents. The desktop becomes a network terminal. Many of a UNIX system's shortcomings are made up by its ease in executing remote commands.
An area of host management that deserves special mention is network initialization. Rather than manually configuring every IBM PC on a network, engineers would rather have the computers use the network to learn their own settings. This way, changing an IP address numbering scheme, for example, would only require changes on a few servers, rather than changing a manual IP address setting on every machine. Furthermore, many users don't have the slightest clue what an IP address is, so the more automated this process can be, the better. Of course, using the network to determine settings needed to use the network presents a chicken and egg problem, making this one of the most un-elegant areas of the Internet protocol suites.
Several schemes have been designed for IP configuration over LANs, permitting hosts to locate their IP addresses, subnet masks, routers and DNS servers without having this information pre-configured. Reverse ARP (RARP), documented in RFC 903, is a simple method to assign IP addresses based on Ethernet addresses. The bootstrap protocol (BootP), documented in RFC 951, is similar, but uses UDP, and can communicate a small amount of additional information. The Dynamic Host Configuration Protocol (DHCP), documented in RFC 2131, is structured as an extension to BootP and can communicate a great deal of additional information, making it the most currently popular solution to LAN host initialization.
The best way to initialize most dial-up serial connections is to use the address negotiation features built in to the PPP Protocol. SLIP users are out of luck, and have to configure addresses manually - this is probably the biggest single factor driving the move to PPP.
In the world of routers, network management is based on the concept of a Management Information Base (MIB), a database of manageable items. In a typical MIB, you will find the system name, IP routing tables, and counts of the packets handled by each protocol. The most popular protocol in use to manipulate MIBs is the Simple Network Management Protocol (SNMP). A second version of SNMP has been developed, with better security, for one thing. SNMPv2 is part of the lofty-sounding Internet-Standard Network Management Framework Version 2, and is accompanied by nearly a dozen attending documents. RFC 1441 introduces the entire "framework".
The MIB model seems to be quite powerful, SNMP is a quality protocol, and dozens of MIBs have been written for all kinds of network devices. Few have been implemented, though, simply because few people can really use any of them. The lack of cheap, good quality user interfaces has seriously hampered the acceptance of SNMP. A complete network management system has to make copy available to the user, as well as /etc/rc to the engineer. SNMP gives us /etc/rc, but nothing for the user. SNMP is invisible to the average user, and will never attain its full potential until simple, easy interfaces are made available to the masses. The Internet business community seems to have decided that SNMP is for engineers, and engineers have money, and the rest of us don't use SNMP.
\end{soapbox}