When to Run a Time Server

One of the primary reasons for running a time server is to deliver accurate time settings to clients on your network. By having clients set their clocks to a time maintained by one of your systems, you can keep your computers' clocks set to reasonably consistent values. (Depending on the protocol in use, "reasonably consistent" can mean variations of just a few milliseconds .) This can avoid problems caused by time mismatches . For instance, with mismatched clocks, a client might save a file to a server, but become confused when reading the file back one minute later because the save date might appear to be two minutes in the future. Mismatched times can also be a real headache when tracking down information in log files. It can also be annoying to see different times on different computers' displays. Some tools, such as Kerberos, rely on clients and servers having consistent times, so running a time server with such tools is a practical necessity.

The time server program that's discussed at greatest length in this chapter is a bit unusual in that it functions as both a server and a client. Therefore, another reason to run this particular time server is to set the clock of the computer on which the server runs. In a network that contains several Linux or UNIX computers, you'll set one machine to obtain its time from an outside server, then configure the rest of the systems in exactly the same way, except that you'll point these systems to the first one to obtain their time locally. This multi-tiered configuration reduces the load on the ultimate sources of time signals, such as computers that are linked to atomic clocks or radios that can receive broadcast time signals from official time sources. If you don't want to run the complete server package on clients, you may be able to make do with a simpler program that functions only as a client, but you'll need to call it explicitly every now and then to keep the system's time set correctly.