Sophie

Sophie

distrib > Mandriva > 2010.1 > x86_64 > media > main-release > by-pkgid > 7e3702e37c5b52c74659c5133cffb3d2 > files > 6

adjtimex-1.29-1mdv2010.1.x86_64.rpm

ADJTIMEX 

I will assume you have already read the man page for adjtimex :-)


BACKGROUND

You probably already know that your computer has two clocks - the "real
time clock" (RTC) that runs all the time, and the system clock that runs
only while the computer is on.  The system clock has much better
precision (approximately 1 usec), but the RTC probably has better
long-term stability.  adjtimex can be used to approximately correct the
system clock for drift.

The handiest reference clock is the RTC.  

	adjtimex --compare

shows the time, the difference between the two clocks, the first
difference (how much the system clock has gained or lost relative to
the RTC since the last check), and two kernel time parameters: tick
and frequency offset.  After the first two comparisons, it also prints
suggested values of tick and frequency offset that will adjust the
system clock to approximately agree with the RTC.  Of course, the RTC
may have a systematic error of its own.  clock(8) or hwclock(8) use
the file /etc/adjtime to record the drift rate and the time since the
RTC was last set.  adjtimex also consults this file and applies the
correction to the times from the RTC.

adjtimex can also record the times on both clocks, and optionally
another reference time, in a log.  It can later compare the logged
time and estimate drift rates.


INSTALLATION

Type "./configure" to configure, "make" to compile, then "make
install" to copy the binary to /usr/local/bin, and the man page to
/usr/local/man/man8.  If you want the program installed in /usr rather
than /usr/local, start with "./configure --prefix=/usr".


ASIDES

You will have to log in as root for most of the following.

The standard kernel will leave the RTC alone only until one of the
time parameters is set (using adjtimex, for example).  Then it thinks
ntpd is regulating the system time, and it starts "correcting" the
RTC approximately every eleven minutes.  adjtimex prevents this by
setting the time after setting any parameter.  It does not actually
change the time.


USAGE

The file /etc/adjtime is used to correct for systematic error in the
RTC.  Set it up as explained in the man page for hwclock(8).  Otherwise,
without /etc/adjtime, adjtimex assumes that the RTC is already perfect.
The latest version of hwclock(8) from Bryan Henderson can be found at
<URL:http://giraffe-data.com/software/about_hwclock.html>.  An older
version is included (along with a lot of other useful utilities) in the
util-linux collection.

I'll explain the manual procedure first.

Compile and run adjtimex with the --compare switch.
Suppose the result is:

	# adjtimex --compare
	794625929   -0.597638 -0.597638  ...
	794625939   -0.602050 -0.004412  ...
			      ^^^^^^^^^
This number says the system clock lost over 4 msec in just 10 sec with
respect to the RTC, or 4 parts in 10000.  Therefore, we increase the
tick parameter to 10004, and repeat the comparison:

	# adjtimex  --tick 10004
	# adjtimex --compare
	794626037   -0.644860 -0.644860  ...
	794626047   -0.645259 -0.000399  ...

Now the system clock is slow by about 39.9 parts per million.  We can
change the frequency offset to fix this.  Recall that the frequency
offset is in parts per million, with the fraction point in the center
(16 bits on either side).  Therefore, we set the offset to 39.9*65536 =
2614886:

	# adjtimex  --frequency 2614886
	# adjtimex --compare
	794626313   -0.652147 -0.652147  ...
	794626323   -0.652247 -0.000100  ...

It still seems to be off by about 10 ppm, so we increase the offset
by 10*65536, to 3270246:

	# adjtimex  --frequency 3270246
	# adjtimex --compare
	794627233   -0.795221 -0.795221  ...
	794627243   -0.795231 -0.000009  ...

The entries in the "suggested" columns are calculated in this way.
Using the --adjust switch automates the process - periodically
installing the "suggested" values.

By the way, using the --offset or --singleshot options seems to have a
long-term effect on the clock.  I suggest you determine the tick and
frequency settings before trying the other options.

You may already have an entry like this in /etc/rc.d/rc.S,
/etc/init.d/boot, or one of the other startup files:

	# Adjust CMOS clock based on /etc/adjtime, then set system clock.
	# Add -u switch if your system keeps GMT.
	if [ -x /sbin/clock ]; then
	  /sbin/clock -a
	fi

Just add an entry like this:

	# Regulate the sytem clock
	if [ -x /sbin/adjtimex ]; then
	  /sbin/adjtimex  --tick 10004  --frequency 3270246 
	fi


LONG TERM COMPARISONS

You can use adjtimex to record the values of the CMOS and system
times, and optionally a reference time, so drift rates can be
determined over long periods.  To enter a reference time manually, use

	# adjtimex  --log --watch

adjtimex will ask you to hit a key when you know the time, then will
ask the time and its approximate precision.  It will then ask whether
either clock has been disturbed since the previous log entry.  You can
take the time from an accurate clock, a radio time broadcast like WWV
or CHU, etc.

If you have ntpdate installed (part of the xntp package), and your
computer has network connectivity, use

	# adjtimex  --log  --host www.xxx.yyy.zzz

and adjtimex will use ntpdate to request the time from the specified
host.  (Ask your Internet Service Provider whether he has an NTP time
server, or can suggest one.)

After you have made several log entries, you can ask adjtimex to
review the log with

	# adjtimex  --review

It will take each pair of log entries in turn, and calculate the error
in each clock's rate in parts per million (ppm), with its estimated
uncertainty.  It will also find over-all average errors, and will
suggest parameter values to correct for drift.  Be sure to check the
outputs for consistency.  One incorrect reference time can throw off
all the calculations.


FURTHER READING

For authoritative documentation on the kernel time software, see
the following, all by David L. Mills <mills@udel.edu>:

[MIL92a] Mills, D.L. Network Time Protocol (Version 3) specification,
implementation and analysis, RFC 1305, University of Delaware, March
1992, 113 pp.

[MIL92c] Mills, D.L. Simple Network Time Protocol (SNTP), RFC 1361,
University of Delaware, August 1992, 10 pp.

[MIL94a] Mills, D.L. A kernel model for precision timekeeping. ARPA
Network Working Group Report RFC-1589, University of Delaware, March
1994. 31 pp.

README.kern, which can be found in louie.udel.edu:/pub/ntp/kernel.tar.Z.


THANKS

to David Mills for the kernel timekeeping code, and Steven Dick
<ssd@nevets.oau.org> for writing adjtimex originally.  (I've changed
the code a lot.  Any problems are my responsibility.)  Thanks also to
Michael Meskes <meskes@debian.org>, whose Debian installation script
inspired the "--compare" and "--adjust" options.


Enjoy!
                - Jim Van Zandt <jrvz at comcast.net>