1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
|
adjtime, tick and tickadj:
--------------------------
The SGI value for HZ is 100 under Irix 4, with the system clock running
in nominal mode (ftimer off), so the value for tick is 10000 usec.
Tickadj is a bit more tricky because of the behaviour of adjtime(),
which seems to try to perform the correction over 100-200 seconds, with
a rate limit of 0.04 secs/sec for large corrections. Corrections of
less than 0.017 seconds generally complete in less than a second,
however.
Some measured rates are as follows:
Delta Rate (sec/sec)
> 1 0.04
0.75 0.04
0.6 0.004
0.5 0.004
0.4 0.0026
0.3 0.0026
0.2 0.0013
0.1 0.0015
0.05 0.0015
0.02 0.0003
0.01 0.015
Strange. Anyway, since adjtime will complete adjustments of less than
17msec in less than a second, whether the fast clock is on or off, I
have used a value of 150usec/tick for the tickadj value.
Fast clock:
-----------
I get smoother timekeeping if I turn on the fast clock, thereby making
the clock tick at 1kHz rather than 100Hz. With the fast clock off, I
see a sawtooth clock offset with an amplitude of 5msec. With it on,
the amplitude drops to 0.5msec (surprise!). This may be a consequence
of having a local reference clock which spits out the time at exactly
one-second intervals - I am probably seeing sampling aliasing between
that and the machine clock. This may all be irrelevant for machines
without a local reference clock. Fiddling with the fast clock doesn't
seem to compromise the above choices for tick and tickadj.
I use the "ftimer" program to switch the fast clock on when the system
goes into multiuser mode, but you can set the "fastclock" flag in
/usr/sysgen/master.d/kernel to have it on by default. See ftimer(1).
timetrim:
---------
Irix has a kernel variable called timetrim which adjusts the system
time increment, effectively trimming the clock frequency. Xntpd could
use this rather than adjtime() to do it's frequency trimming, but I
haven't the time to explore this. There is a utility program,
"timetrim", in the util directory which allows manipulation of the
timetrim value in both SGI and xntpd native units. You can fiddle with
default timetrim value in /usr/sysgen/master.d/kernel, but I think
that's ugly. I just use xntpd to figure out the right value for
timetrim for a particular CPU and then set it using "timetrim" when
going to multiuser mode.
Serial I/O latency:
-------------------
If you use a local clock on an RS-232 line, look into the kernel
configuration stuff with regard to improving the input latency (check
out /usr/sysgen/master.d/[sduart|cdsio]). I have a Kinemetrics OM-DC
hooked onto /dev/ttyd2 (the second CPU board RS-232 port) on an SGI
Crimson, and setting the duart_rsrv_duration flag to 0 improves things
a bit.
12 Jan 93
Steve Clift, CSIRO Marine Labs, Hobart, Australia (clift@ml.csiro.au)
|
