Re: [ntp:questions] Q: Disabling "11 minute mode"

2008-02-28 Thread Ulrich Windl
[EMAIL PROTECTED] (Dean S. Messing) writes:

> Serge Bets wrote:
>>  On Tuesday, January 22, 2008 at 19:02:23 +, Dean S. Messing wrote:
>> 
>> > Is it possible to disable "11 minute mode" from "ntp.conf"?
>> 
>> No. You have to tweak the kernel. If you have the PPSkit:
>> 
>> | $ echo 0 > /proc/sys/kernel/time/rtc_update
>> 
[...]
> Thanks Serge.  I looked up PPSkit.  Looks good, but I'm going to have
> to learn how to patch the Fedora kernel to install PPSkit.

PPSkit is based on kernel 2.4. Recent changes in kernel 2.6 were dramatic
regarding the time code. I'm unsure if an equivalent mechanism exists in 2.6,
but it would be nice if someone would port "the pearls" to kernel 2.6.

[...]

Ulrich

___
questions mailing list
questions@lists.ntp.org
https://lists.ntp.org/mailman/listinfo/questions


Re: [ntp:questions] Q: Disabling "11 minute mode"

2008-01-25 Thread Dean S. Messing

Serge Bets wrote:
>  On Tuesday, January 22, 2008 at 19:02:23 +, Dean S. Messing wrote:
> 
> > Is it possible to disable "11 minute mode" from "ntp.conf"?
> 
> No. You have to tweak the kernel. If you have the PPSkit:
> 
> | $ echo 0 > /proc/sys/kernel/time/rtc_update
> 
> Otherwise you have to patch time.c in the kernel. Dead easy, just a
> matter of commenting out a line or two. I'm so patching all my kernels,
> reading and writing the RTC exclusively with hwclock 2.31, and am
> getting a far better accuracy.
> 
> The main purpose of an RTC is to initialise the system time at powerup,
> isn't it? Most people startup in the morning at around half a second of
> the true time, and later ntpd has to step this to UTC. I routinely
> startup at some low milliseconds of the true time, offset quickly
> slewed. My last step event was years ago.

Thanks Serge.  I looked up PPSkit.  Looks good, but I'm going to have
to learn how to patch the Fedora kernel to install PPSkit.

But I'm discovering that I have rather deeper problems on my machine
(a Dell 490 Precision).  Using adjtimex --compare to track the drift
between system and cmos clock (ntpd not running), I see that the RTC
is behaving _very_ strangely.  It will begin to return screwy values
after several hours of doing "adjtimex --compare" and then get to the
point where "hwclcok --show" hangs.  So my desire to turn off "11
minute mode" is  mute when ntp is running is mute.

For your amusement, here's a snippet of the output of
"adjtime --compare" with an interval of 60 seconds:

1200982902 0.001784   -2.0  10001   3929312   10001   4060301
1200982962 0.0017920.1  10001   3929312   10001   3920719
1200983022 0.0020514.3  10001   3929312   10001   3646240
1200983082 0.001828   -3.7  10001   3929312   10001   4173062
1200983142 0.001756   -1.2  10001   3929312   10001   4007957
1200983202 0.0020254.5  10001   3926656   10001   3632906
1200983261 0.500370 8305.8  10001   39262889918   3549307
120098328140.001689   658355.3  10001   39262883418301130
120098334140.0019314.0  10001   3926288   10001   3661966
120098340734.001894  -10.6  10001   3926288   11001   3966652
120098346140.0016465.9  10001   39262889001   4197121
120098352140.0018904.1  10001   3926288   10001   3659882
120098360912.001763  -48.8  10001   3924640   14668   1878649
120098364140.001606   44.0  10001   39246405334   6280787
1200983741 0.001726  -64.7  10001   3924640   16668   1609118
120098376140.001911   69.8  10001   39246403334   5907090
120098382140.001553   -6.0  10001   3924640   10001   4315525
1200983921 0.001748  -63.4  10001   3924640   16668   1527086
120098394140.001894   69.1  10001   39246403334   5949798
120098400140.001554   -5.7  10001   3924640   10001   4295994
1200984101 0.001700  -64.2  10001   3921488   16668   1577580
1200984161 0.001291   -6.8  10001   3921104   10001   4367718
1200984221 0.0015324.0  10001   3921104   10001   3657823
1200984275 6.001806   14.6  10001   39211049001   3621886
120098430140.001722   55.3  10001   39203684334   6196568
120098436140.0019744.2  10001   3920368   10001   3645108
120098442734.001868  -11.8  10001   3920368   11001   4036253
120098448140.0016796.8  10001   39203689001   4126878

Things got so bad that the output eventually became:

199345540  1001658696.064552  1592732.9  10001   3879376   -5926   1725431
199345717  1001658600.500585 -1592732.8  10001   3879376   25928   6027853
199345718  1001658696.023830  1592054.1  10001   3879376   -5919335126
199345896  1001658600.500586 -1592054.1  10001   3879376   25922868985
199345897  1001658696.045414  1592413.8  10001   3879376   -5923   2975047


Before it went crazy, it had run smoothly for 5 or 6 hours.
When I rebooted into the BIOS and looked at the RTC it was off by
several years.

This has now happened thrice, but only when adjtimex is running in the
compare mode for long periods.  I have no idea what this means.  The
cmos battery does not appear to be the problem since, after a reboot,
the RTC remains at proper time indefinitely (modulo drift), unless
and until I run adjtimex --compare for several hours.

Anyway, thanks for the info. on "11 minute mode".  Wish I could fix my
RTC problem

Dean
___
questions mailing list
questions@lists.ntp.org
https://lists.ntp.org/mailman/listinfo/questions


Re: [ntp:questions] Q: Disabling "11 minute mode"

2008-01-25 Thread Dean S. Messing

Serge Bets wrote:
:  On Tuesday, January 22, 2008 at 23:13:22 +, Dean S. Messing wrote:
: 
: > do "adjtimex -p" and look at the "status:" value. If it's odd,
: > (LSB==1) then your kernel is in "11 minute mode".
: 
: Not exactly: bit #0 set means your kernel is in PLL mode. That's bit #6
: unset that means eleven-minutes mode. The kernel does not always enter
: in both modes together.

Correct.  Thanks for correcting this bit of mis-information.  I was
mis-remembering the fact that by disabling PLL mode (with the ntp
"disable kernel" directive, I think) _and_ setting status bit #6,
I could keep the kernel out of "11 minute mode".  That was done a
couple of years ago on another machine and I was confusing the two
things.

Dean
___
questions mailing list
questions@lists.ntp.org
https://lists.ntp.org/mailman/listinfo/questions


Re: [ntp:questions] Q: Disabling "11 minute mode"

2008-01-25 Thread Serge Bets
 On Tuesday, January 22, 2008 at 23:13:22 +, Dean S. Messing wrote:

> do "adjtimex -p" and look at the "status:" value. If it's odd,
> (LSB==1) then your kernel is in "11 minute mode".

Not exactly: bit #0 set means your kernel is in PLL mode. That's bit #6
unset that means eleven-minutes mode. The kernel does not always enter
in both modes together.


Serge.
-- 
Serge point Bets arobase laposte point net

___
questions mailing list
questions@lists.ntp.org
https://lists.ntp.org/mailman/listinfo/questions


Re: [ntp:questions] Q: Disabling "11 minute mode"

2008-01-25 Thread Serge Bets
Hello Dean,

 On Tuesday, January 22, 2008 at 19:02:23 +, Dean S. Messing wrote:

> Is it possible to disable "11 minute mode" from "ntp.conf"?

No. You have to tweak the kernel. If you have the PPSkit:

| $ echo 0 > /proc/sys/kernel/time/rtc_update

Otherwise you have to patch time.c in the kernel. Dead easy, just a
matter of commenting out a line or two. I'm so patching all my kernels,
reading and writing the RTC exclusively with hwclock 2.31, and am
getting a far better accuracy.

The main purpose of an RTC is to initialise the system time at powerup,
isn't it? Most people startup in the morning at around half a second of
the true time, and later ntpd has to step this to UTC. I routinely
startup at some low milliseconds of the true time, offset quickly
slewed. My last step event was years ago.


Serge.
-- 
Serge point Bets arobase laposte point net

___
questions mailing list
questions@lists.ntp.org
https://lists.ntp.org/mailman/listinfo/questions


Re: [ntp:questions] Q: Disabling "11 minute mode"

2008-01-22 Thread Unruh
David Woolley <[EMAIL PROTECTED]> writes:

>Richard B. Gilbert wrote:

>> 
>> What IS "11 minute mode"??

>If the linux kernel is told that its clock is being disciplined, it sets 
>the RTC from the software clock every 11 minutes.  Most Unices do 
>something like this, although sometimes it is a cron job.

Yes, it is a horrible wayto discipline the rtc. 
Better to measure the offset and drift and use that info when you read the
rtc. 
You have to tell the kernel that the clock is unsynchronized ( the status
byte in struct timex-- TIME_BAD
If that is set, the 11 min mode is switched off. 
But I think youhave to keep doing it, since the kernel may decide its
status is fine. 


___
questions mailing list
questions@lists.ntp.org
https://lists.ntp.org/mailman/listinfo/questions


Re: [ntp:questions] Q: Disabling "11 minute mode"

2008-01-22 Thread Dean S. Messing

Richard B. Gilbert wrote:
> Dean S. Messing wrote:
> > Is it possible to disable "11 minute mode" from "ntp.conf"?
> > I've tried using the command "disable kernel" but that
> > appears to change the way time discipline is maintained, but
> > does nothing for "11 minute mode".
> > 
> > If using ntp.conf is not the way, what is?
> > 
> > Thanks for your help.
> > Dean
> 
> What IS "11 minute mode"??

Oops.  Sorry!  I thought everyone reading this list
(who could answer my question :-) would know.

David Woolley gave you a good answer already so
I'll only add that if you want to "see" if you
are in "11 minute mode", do "adjtimex -p" and
look at the "status:" value. If it's odd, (LSB==1)
then your kernel is in "11 minute mode".

Now, if someone would tell me how to disable it
(short of hacking time.c) I'd be most thankful.

I tried turning it off with adjtimex -S 64) but
ntp changes it back again in a few minutes.

I'd like to disable it, but keep ntp kernel
discipline so I can do some analysis of my RTC.

Dean
___
questions mailing list
questions@lists.ntp.org
https://lists.ntp.org/mailman/listinfo/questions


Re: [ntp:questions] Q: Disabling "11 minute mode"

2008-01-22 Thread David Woolley
Richard B. Gilbert wrote:

> 
> What IS "11 minute mode"??

If the linux kernel is told that its clock is being disciplined, it sets 
the RTC from the software clock every 11 minutes.  Most Unices do 
something like this, although sometimes it is a cron job.

___
questions mailing list
questions@lists.ntp.org
https://lists.ntp.org/mailman/listinfo/questions


Re: [ntp:questions] Q: Disabling "11 minute mode"

2008-01-22 Thread Richard B. Gilbert
Dean S. Messing wrote:
> Is it possible to disable "11 minute mode" from "ntp.conf"?
> I've tried using the command "disable kernel" but that
> appears to change the way time discipline is maintained, but
> does nothing for "11 minute mode".
> 
> If using ntp.conf is not the way, what is?
> 
> Thanks for your help.
> Dean

What IS "11 minute mode"??


___
questions mailing list
questions@lists.ntp.org
https://lists.ntp.org/mailman/listinfo/questions


[ntp:questions] Q: Disabling "11 minute mode"

2008-01-22 Thread Dean S. Messing

Is it possible to disable "11 minute mode" from "ntp.conf"?
I've tried using the command "disable kernel" but that
appears to change the way time discipline is maintained, but
does nothing for "11 minute mode".

If using ntp.conf is not the way, what is?

Thanks for your help.
Dean
___
questions mailing list
questions@lists.ntp.org
https://lists.ntp.org/mailman/listinfo/questions