Re: NTP dumps Linux, film at 11. [Fwd/FYI]
Jeffrey Hutzelman (jhutz+@cmu.edu)
Wed, 2 Dec 1998 20:04:08 -0500 (EST)
> In message <199812020115.UAA29332@dcl.MIT.EDU>, "Theodore Y. Ts'o" writes:
> +-----
> | Date: Tue, 01 Dec 1998 18:03:07 -0500
> | From: "Brandon S. Allbery KF8NH" <allbery@kf8nh.apk.net>
> |
> | The other thing about it is that it's not just lost interrupts: the
> | systems are as apt to *gain* time, instead of losing it, again enough
> | to make Kerberos upset (I've seen them gain 10 minutes over 5 minutes
> | of wall time). Again, I haven't been able to correlate gain vs. loss
> | (or amount thereof) to any particular program or device/driver.
> |
> | Hmm... I've never seen this myself. I've only lost time, on my clock,
> | and only under very explainable circumstances (such as using the
> | Bustoaster SCSI card). And I use Kerberos all the time, so I tend to
> +--->8
>
> It was very noticeable. I upgraded the entire cluster (14 machines) in
> HH1202 one afternoon; half the machines started gaining time like mad, the
> other half lost time like mad. All machines identical, indeed from the
> same manufacturing batch as far as I can tell (and all idle, the spring
> semester having just ended at the time). Within 15 minutes all of them
> were outside of the time window the AFS kaservers were willing to grant
> them (the usual 5 minutes for Kerberos) and linux-afs had lost the battle
> to resync their clocks. (10 minutes later xntpd was a standard part of our
> Linux install....)
>
> Later in the summer I upgraded them all to RH5.1, with xntpd disabled
> initially so I could see how they would behave. They proceeded to do much
> the same thing, except that a few of them waited most of a week before
> suddenly beginning to drift wildly.
>
> Every (RH5-based) Linux system I've encountered, whether installed by me or
> by someone else, has behaved similarly, although not always with such a
> massive time slew (it varies widely and for no discernable reason, even on
> the same machine).
>
> I've only installed 5.2 on a handful of machines so far, but the time
> slippages are still there so far.
I've _never_ seen this behaviour or had users report it; perhaps there's
something unusual about your installs that's making it happen, or about
ours that makes it go away. We're presently using ntpd and running AFS
with -nosettime, as God intended. FWIW, I'm pretty sure that Andrew uses
the time-sync support in AFS on their 4.2-based workstations (as they do
on every other platform), so I'm pretty sure that code works correctly.
We, of course, have the same situation as you or Ted - if our machines'
time starts to fluctuate wildly, the users notice pretty quickly, because
bad things happen like people not being able to log in...
-- Jeffrey T. Hutzelman (N3NHS) <jhutz+@cmu.edu>
Systems Programmer
School of Computer Science - Research Computing Facility
Carnegie Mellon University - Pittsburgh, PA
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/