davie said:
davie wrote on 05/09/2011 06:57 ET :
I think I need to expand on the description of this problem. We are in a
bunker
style enviroment so we cannot use a gps ntp solution.
The most important thing for me is that I achieve uniformity accross the
network
so that all the computers refer to one computer. If that computer is off by 1
or
2 minutes is not important. It is more important that they are all the same.
Is
there a solution to this?
Sure, make one of the computers a server, by finding an "ntpd" program for it.
The other computers, run ntp client processes, and get their synchronization
from the master computer.
You don't need to run GPS, but GPS is a cheap way of doing it, since a
GPS receiver doesn't have to cost a lot.
There are various other kinds of stand alone oscillators, which have slightly better
characteristics than the computer. One benefit of such schemes (taking the form
perhaps, of a plug in card for the computer), is that they are "pure hardware",
so the actual time keeping process isn't interfered with by software.
The problem with some computers, is their local time keeping isn't that
good. And that has to do with the usage of "counting clock tick interrupts"
as an OS level time keeping mechanism. If there are no issues with the
hardware or software, that can give adequate performance. But on a
buggy machine, using such a machine for a clock source - you'd be better
off with a sundial.
So before you nominate a computer, you'd want to make sure it has reliable
characteristics, and doesn't have large random drifts that don't make sense.
Otherwise, none of the computers will end up with good time keeping.
If you can't be bothered to evaluate your master computer for its drift
performance, then pop in a card. It might cost you up to $1000 for one
of those, depending on how much time you have to shop around. This one,
for example, is a cheap digital watch style implementation, using a
32768 Hz watch quartz crystal. It claims "2 seconds/month" in a temperature
stable environment, which is a lot better than any computer can manage
on its own. A cesium or rubidium based device would cost a lot more, and
give much better results (which you probably don't care about).
http://www.atomictimeclock.com/clcaspecs.htm
http://www.atomictimeclock.com/mainspecials.htm
ClockCard PCI Server Reg: $279.95 Sale $249.95 (save $30)
http://www.atomictimeclock.com/client_server.php
I included a link for their client/server software, but if
your budget is tight, you just might be able to find something
free to use. You'd think they'd throw the server software in
with the card
If the software followed any standards at
all, you'd think the w32time client on the client PCs would be
good enough.
Paul