Path: utzoo!attcan!uunet!ncrlnk!ncr-sd!hp-sdd!hplabs!hpfcdc!hpldola!hp-lsd!prisma!mo
From: mo@prisma
Newsgroups: comp.periphs
Subject: Re: precise time for systems?
Message-ID: <8100001@prisma>
Date: 25 Dec 88 17:42:00 GMT
References: <460@n3dmc>
Lines: 28

Dave Mills has done a lot of work developing network time synchronization
protocols.  It turns out the limiting factor is the variance of the transmission
delays - the not the size of the delay itself.  There are several RFC's
and some code for 4.3BSD implementing the Network Time Protocol.  The scheme
is based on a heirarchy of clocks with the assumption that some hosts will
have WWV time code receivers.  The current system has both "professional"
receivers and Heathkit "Most Accurate Clocks".  They work well and on
a local network you can keep the time between machines synced to 
microsecond timescales, while millisecond accuracy is possible across
some Internet paths.

Most interstingly, monitoring the time drifts are spectacular ways of
doing remote measurement.  For instance, if your local clock is crystal-based
and not in an oven, Dave can, over a period of weeks, calibrate your
machine room and tell you the temperature by watching the temp-induced
time drift.  He has also looked at the synchronization of the national
power grid.  Once it was all commonly synced, but not any more.  There are
regional grids which drift relative to one another.  Further, he discovered
that by watching line-time clocks that 10 minutes before the evening half hours,
in Britian everyone goes and puts the electric kettle on for tea.  THhe
sudden surge dips the line frequency and the clocks drift faster. He finally
asked someone about it and it turns out that is when the commercials come on...

SO if you are interested in distributed timing, become a clock ticker yourself
and read Dave's stuff before setting out on your own...

	Yours for reinventing flat tires,
	-Mike O'Dell