[3509] in testers
Re: sun4 [8.2.0]: time syncronization
daemon@ATHENA.MIT.EDU (John Hawkinson)
Wed Jun 3 10:43:02 1998
To: "Richard Basch" <basch@lehman.com>, testers@MIT.EDU
In-Reply-To: Richard's message of "Wed, 03 Jun 1998 10:11:23 EDT."
<199806031411.KAA14699@badger.lehman.com>
Date: Wed, 03 Jun 1998 10:42:51 EDT
From: John Hawkinson <jhawk@MIT.EDU>
In message <199806031411.KAA14699@badger.lehman.com>, basch@lehman.com writes:
>Are you running Solaris 2.6 and NTP?
Sigh; I knew I was missing something important in my analysis of this problem
(I haven't actually played with any 8.2 machines yet, but I've been scratching
my head pretty hard at the reports). The important part being it hadn't clicked
for me that this was 2.6.
Yeah.
>I had problems at Lehman. In fact, I had to give up on the PD NTP
>source and use Sun's NTP. Even then, I found that there was quite a bit
>of clock drift and the Suns sometimes would not sync. I have also heard
>that Sun has messed up the PLL code in the kernel in 2.6, but I cannot
>confirm that.
According to the NTP FAQ,
http://www.eecis.udel.edu/~ntp/database/FAQ/
which is hopelessly disorganized to the extent that it's usually easier to use
DejaNews to find information, Sun added support for kernel PLL to Solaris 2.6,
but broke it in the process.
http://www.eecis.udel.edu/~ntp/database/FAQ/faq3Carn.htm#arana
is a URL that sort of gets one going.
The <cough> <sputter> <wheeze> "recommended" fix appears to be to cmment out
'#define KERNEL_PLL 1' from config.h after configure runs, though I'm sure
something better can be done.
Supposedly Sun's ntp works just fine because it knows not to use kernel
PLL. According to Casper Dik, this is Sun BugID 4095849. Some other text
implies that there may be a problem with some hardware clock drivers, but I
don't think that's really relevent to
us. (c.f. "http://www.eecis.udel.edu/~ntp/database/FAQ/faq3Carm.htm#aramu").
Inspection of 4095849 on SunSolve indicates that it's certainly relevent
to NTP and suggests you disable KERNEL_PLL (even in the aforementioned crude
fashion), but the symptom of the problem is somewhat different, viz. the PLL code
can cause panics under some circumstances.
So it's hardly 100% clear that removing KERNEL_PLL will fix the problem, but it
at least seems likely.
It is with some irony and some trepidation that I note that upon Fri, 12 Sep
1997 04:16:03 -0400, I told Dave Mills I'd try and find some time to update the
file in the NTP source distribution that talks about Solaris-specific stuff,
and still haven't quite gotten around to it. As such, I suppose it is unfair
for me to complain overmuch about the underdocumented nature of this.
Lastly, I think it someone should consider whether we should bump the ntp
version from xntpd 3.51 to ntp 4.0.72j; despite the fact that the major version
bumped from 3 to 4 and Dave finally dropped the "x" (which stood for, "experimental",
'ya know), I don't think there have been huge changes.
On the other hand, the current 3.x release is 5.93 of 27 April 1998.
Thanks, Richard.
--jhawk
P.S. releng, did I miss something such that there's a good reason why
the most recent wash status file in AFS is dated 24 April?
[contents-vnder-pressvre!jhawk] /afs/dev/service/wash> ls -l /afs/dev/service/wash/status/kenmore
total 3188
-rw-r--r-- 1 33554434 root 1612831 Apr 24 09:53 current
-rw-rw-r-- 1 33554434 root 17957 Apr 24 09:53 logfile
