If this program is run on the local servers, their clocks can be kept synchronized; other machines can then use the server(s) as a time standard.

This program is a lot more efficient than Sun's tool, but it isn't really good for better than the nearest second. This suffices for editing and make.

Options start with '+' (enable) or '-' (disable), followed by a letter, and possibly a value. The case of the option letter is not significant, though the case of letters in values is usually significant.

-a

Don't do the adjustment. Default: +a.

+a

Do the adjustment. Will fail if not super-user.

-d<l><file>

+d<l><file>

Debug mode: <l> is 1-digit level; <file> is output file. Default is "+d1", fatal errors only, written to stderr.

-f

+f<n>

Repeat forever. The -f and -r option suppress this; only one syncs will be done, and then we exit. The +f and +r option mean to keep running and repeat the operation every <n> seconds. Default: -f, which will bring the clock into sync and exit, after which the local clock may slowly drift..

-i

+i<n>

Interval between repeats, in seconds. The "-i" option suppresses repeats, and is equivalent to "-f". The default is "-i". If you use +i, <n> should probably be several hours, unless your clock is particularly bad.

-r

+r<n>

Use the rtime() routine to query the authorities. If <n> is given, it is the interval between queries. The default is "-r". This is not used by default, because few systems seem to have this service running. If it is common at your site, you might want to change the default to use this rather than +t.

-t

+t<n>

Use the UDP time service to query the authorities. <n> is the UDP port to use; the default is "+t37".

-v<l><file>

+v<l><file>

Verbose mode; the same as the 'd' debug option.

The default behavior with only a list of authority hostnames is to use the UDP time service just once, wait up to 5 seconds for replies, and then use the mean time in the replies to adjust the system clock. The result will be clocks that are synchronized well enough (within a second) for the purpose of things like make(1), but not nearly well enough for, say, long-baseline interferometry.

There are some machines on the internet, mostly IBM mainframes, whose time service believes that 1900 was a leap year. Such machines will give a time that is off by one day. Don't use them. And send in a bug report. This will probably be a problem after 2000/2/29, which many machines will consider to be 2000/3/1, but this program shouldn't be affected, since we use only their binary timestamps.