xref: /freebsd/contrib/ntp/html/hints/solaris.html (revision 2abb9b42a53c39a8b781e25a05de1de9e98b8b9a)
1<HTML>
2<HEAD>
3<TITLE>Solaris hints and kinks</title><link href="scripts/style.css" type="text/css" rel="stylesheet">
4
5</HEAD>
6<BODY>
7Information on compiling and executing ntpd under Solaris.
8<BR>
9<p>Last update:
10  <!-- #BeginDate format:En2m -->27-Jan-2014  05:31<!-- #EndDate -->
11  UTC,
12John Hawkinson,
13<! -- This is deliberately not a mailto -- > &lt;jhawk@MIT.EDU&gt;
14</p>
15<P>
16If you're not running Solaris 2.5.1 or later, it is likely
17that you will have problems; upgrading would be a really good plan.
18<P>
19<H3>All Solaris versions</H3>
20<P>
21      We have a report that says starting with Solaris 2.6 we should leave
22      <I>dosynctodr</I> alone.
23      <A HREF="solaris-dosynctodr.html">Here is the report</A>.
24<P>
25Proper operation of ntp under Solaris may require setting the kernel
26variable <I>dosynctodr</I> to zero (meaning "do not synchronize the clock
27to the hardware time-of-day clock").  This can be done with the
28tickadj utility:
29<BLOCKQUOTE><TT>
30tickadj -s
31</TT></BLOCKQUOTE>
32If you prefer, it can also be done with the native Solaris kernel debugger:
33<BLOCKQUOTE><TT>
34echo dosynctodr/W0 | adb -k -w /dev/ksyms /dev/mem
35</BLOCKQUOTE></TT>
36<P>
37Or, it can also be set by adding a line to /etc/system:
38<BLOCKQUOTE><TT>
39set dosynctodr = 0
40</BLOCKQUOTE></TT>
41<P>
42Instead of the <I>tick</I> kernel variable, which many operating
43systems use to control microseconds added to the system time every
44clock tick (c.f. <A HREF="#frequency_tolerance">Dealing
45with Frequency Tolerance Violations</A>), Solaris has the variables
46<I>nsec_per_tick</I> and <I>usec_per_tick</I>.
47<P>
48<I>nsec_per_tick</I> and <I>usec_per_tick</I> control the number of
49nanoseconds and microseconds, respectively, added to the system clock
50each clock interrupt. Enterprising souls may set these based on
51information collected by ntpd in the <CODE>/etc/ntp.drift</CODE> file
52to correct for individual hardware variations.
53<P>
54On UltraSPARC systems, <I>nsec_per_tick</I> and <I>usec_per_tick</I>
55are ignored in favor of the <I>cpu_tick_freq</I> variable, which
56should be automatically be determined by the PROM in an accurate
57fashion.
58<P>
59In general, the same ntp binaries should not be used across multiple
60operating system releases. There is enough variation in the core operating
61system support for timekeeping that a rebuild of ntpd for the idiosyncracies
62of your specific operating system version is advisable.
63<P>
64It is recommended that ntp be started via a script like <A
65HREF="solaris.xtra.S99ntpd">this one</A>, installed in
66<CODE>/etc/init.d/ntpd</CODE> with a symbol link from
67<CODE>/etc/rc2.d/S99ntpd</CODE>.
68
69<a id="frequency_tolerance" />
70<h4>Dealing with Frequency Tolerance Violations (<tt>tickadj</tt> and
71Friends)</h4>
72    The NTP Version 3 specification RFC-1305 calls for a maximum
73    oscillator frequency tolerance of +-100 parts-per-million (PPM), which is
74representative of those components suitable for use in relatively
75inexpensive workstation platforms. For those platforms meeting this
76tolerance, NTP will automatically compensate for the frequency errors of the
77individual oscillator and no further adjustments are required, either to the
78configuration file or to various kernel variables. For the NTP Version 4
79release, this tolerance has been increased to +-500 PPM.  <p>However, in the
80case of certain notorious platforms, in particular Sun 4.1.1 systems, the
81performance can be improved by adjusting the values of certain kernel
82variables; in particular, <tt>tick</tt> and <tt>tickadj</tt>.  The variable
83<tt>tick</tt> is the increment in microseconds added to the system time on
84each interval- timer interrupt, while the variable <tt>tickadj</tt> is used
85by the time adjustment code as a slew rate, in microseconds per tick. When
86the time is being adjusted via a call to the system routine
87<tt>adjtime()</tt>, the kernel increases or reduces tick by <tt>tickadj</tt>
88microseconds per tick until the specified adjustment has been
89completed. Unfortunately, in most Unix implementations the tick increment
90must be either zero or plus/minus exactly <tt>tickadj</tt> microseconds,
91meaning that adjustments are truncated to be an integral multiple of
92<tt>tickadj</tt> (this latter behaviour is a misfeature, and is the only
93reason the <tt>tickadj</tt> code needs to concern itself with the internal
94implementation of <tt>tickadj</tt> at all). In addition, the stock Unix
95implementation considers it an error to request another adjustment before a
96prior one has completed.</p> <p>Thus, to make very sure it avoids problems
97related to the roundoff, the <tt>tickadj</tt> program can be used to adjust
98the values of <tt>tick</tt> and <tt>tickadj</tt>. This ensures that all
99adjustments given to <tt>adjtime()</tt> are an even multiple of
100<tt>tickadj</tt> microseconds and computes the largest adjustment that can
101be completed in the adjustment interval (using both the value of
102<tt>tick</tt> and the value of <tt>tickadj</tt>) so it can avoid exceeding
103this limit. It is important to note that not all systems will allow
104inspection or modification of kernel variables other than at system build
105time. It is also important to know that, with the current NTP tolerances, it
106is rarely necessary to make these changes, but in many cases they will
107substantially improve the general accuracy of the time service.</p>
108<p>Unfortunately, the value of <tt>tickadj</tt> set by default is almost
109always too large for <tt>ntpd</tt>. NTP operates by continuously making
110small adjustments to the clock, usually at one-second intervals. If
111<tt>tickaj</tt> is set too large, the adjustments will disappear in the
112roundoff; while, if <tt>tickadj</tt> is too small, NTP will have difficulty
113if it needs to make an occasional large adjustment. While the daemon itself
114will read the kernel's values of these variables, it will not change the
115values, even if they are unsuitable. You must do this yourself before the
116daemon is started using the <tt>tickadj</tt> program included in the
117<tt>./util</tt> directory of the distribution. Note that the latter program
118will also compute an optimal value of <tt>tickadj</tt> for NTP use based on
119the kernel's value of <tt>tick</tt>.</p> <p>The <tt>tickadj</tt> program can
120reset several other kernel variables if asked. It can change the value of
121<tt>tick</tt> if asked. This is handy to compensate for kernel bugs which
122cause the clock to run with a very large frequency error, as with SunOS
1234.1.1 systems. It can also be used to set the value of the kernel
124<tt>dosynctodr</tt> variable to zero. This variable controls whether to
125synchronize the system clock to the time-of-day clock, something you really
126don't want to be happen when <tt>ntpd</tt> is trying to keep it under
127control. In some systems, such as recent Sun Solaris kernels, the
128<tt>dosynctodr</tt > variable is the only one that can be changed by the
129<tt>tickadj</tt> program.  In this and other modern kernels, it is not
130necessary to change the other variables in any case.</p>
131
132<p>We have a report that says starting with Solaris 2.6 we should leave
133<i>dosynctodr</i> alone.</p> <p>In order to maintain reasonable correctness
134bounds, as well as reasonably good accuracy with acceptable polling
135intervals, <tt>ntpd</tt> will complain if the frequency error is greater
136than 500 PPM. For machines with a value of <tt>tick</tt> in the 10-ms range,
137a change of one in the value of <tt>tick</tt> will change the frequency by
138about 100 PPM. In order to determine the value of <tt>tick</tt> for a
139particular CPU, disconnect the machine from all source s of time
140(<tt>dosynctodr</tt> = 0) and record its actual time compared to an outside
141source (eyeball-and-wristwatch will do) over a day or more. Multiply the
142time change over the day by 0.116 and add or subtract the result to tick,
143depending on whether the CPU is fast or slow. An example call to
144<tt>tickadj</tt> useful on SunOS 4.1.1 is:</p>
145                <pre>
146     <tt>tickadj</tt> -t 9999 -a 5 -s
147</pre>
148which sets tick 100 PPM fast, <tt>tickadj</tt> to 5 microseconds and turns
149off the clock/calendar chip fiddle. This line can be added to the <tt
150>rc.local</tt> configuration file to automatically set the kernel variables
151at boot time.  <p>All this stuff about diddling kernel variables so the NTP
152daemon will work is really silly. If vendors would ship machines with clocks
153that kept reasonable time and would make their <tt>adjtime()</tt> system
154call apply the slew it is given exactly, independent of the value of
155<tt>tickadj</tt>, all this could go away. This is in fact the case on many
156current Unix systems.</p>
157
158<H3>Solaris 2.6</H3>
159<P>
160Solaris 2.6 adds support for kernel PLL timekeeping, but breaks this
161support in such a fashion that using it worse than not. This is <A
162HREF="solaris.xtra.4095849"> SUN Bug ID 4095849</A>, and it is not yet
163fixed as of June 1998.
164<P>
165<H3>Solaris 2.5 and 2.5.1</H3>
166<P>
167On UltraSPARC systems, calculation of <I>cpu_tick_freq</I> is broken
168such that values that are off by significant amounts may be used
169instead. This unfortunately means that ntpd may have severe problems
170keeping synchronization. This is <A HREF="solaris.xtra.4023118"> SUN Bug ID
1714023118</A>. Bryan Cantrill <! -- &lt;bmc@eng.sun.com&gt; --> of Sun
172posted <A HREF="solaris.xtra.patchfreq">patchfreq</A>, a workaround script,
173to comp.protocols.time.ntp in March of 1997.
174<P>
175<HR>
176<H2>OLD DATA</H2>
177<STRONG>I can't vouch for the accuracy the information below this
178rule. It may be significantly dated or incorrect.</STRONG>
179<P>
180<P>
181<H3>Solaris 2.2</H3>
182<P>
183Solaris 2.2 and later contain completely re-written clock code to
184provide high resolution microsecond timers. A benefit of the
185re-written clock code is that adjtime does not round off its
186adjustments, so ntp does not have to compensate for this
187rounding. Under Solaris 2.2 and later, ntp #define's
188<CODE>ADJTIME_IS_ACCURATE</CODE>, and does not look for the <I>tickadj</I>
189kernel variable.
190<P>
191<H3>Solaris 2.1</H3>
192(This originally written by William L. Jones &lt;jones@chpc.utexas.edu&gt;)
193<P>
194Solaris 2.1 contains fairly traditional clock code, with <I>tick</I>
195and <I>tickadj</I>.
196<P>
197Since settimeofday under Solaris 2.1 only sets the seconds part of timeval
198care must be used in starting xntpd.  I suggest the following start
199up script:
200<BLOCKQUOTE><TT>
201tickadj -s -a 1000
202<BR>ntpdate -v server1 server2
203<BR>sleep 20
204<BR>ntpdate -v server1 server2
205<BR>sleep 20
206<BR>tickadj -a 200
207<BR>xntpd
208</TT></BLOCKQUOTE>
209
210The first tickadj turns of the time of day clock and sets the tick
211adjust value to 1 millisecond.  This will insure that an adjtime value
212of at most 2 seconds will complete in 20 seconds.
213<P>
214The first ntpdate will set the time to within two seconds
215using settimeofday or it will adjust time using adjtime.
216<P>
217The first sleep insures the adjtime has completed for the first ntpdate.
218<P>
219The second ntpdate will use adjtime to set the time of day since the
220clock should be within 2 seconds of the correct time.
221<P>
222The second tickadj set the tick adjust system value to 5 microseconds.
223<P>
224The second sleeps insure that adjtime will complete before starting
225the next xntpd.
226<P>
227I tried running with a tickadj of 5 microseconds with out much success.
228200 microseconds seems to work well.
229<P>
230<HR>
231Prior versions of this file had major text contributed by:
232<MENU>
233<LI>Denny Gentry &lt;denny@eng.sun.com&gt;
234</MENU>
235<BODY>
236</HTML>
237