[mythtv-users] Fedora 7 clock drift

[f-myth-users at media.mit.edu]

    > Date: Mon, 08 Oct 2007 14:40:27 +0100
    > From: Mike Perkins <mikep at randomtraveller.org.uk>

    > Jerome Yuzyk wrote:
    > > I am wondering if any Myth users have been seeing clock-drift with Fedora 7. I 
    > > was wondering what was going on when my evening recordings began to start 
    > > noticeably late. For years I have used ntpdate to sync my clock at 1 minute 
    > > after 4am and log the result. On my Myth box pre-Fedora 7 the time difference 
    > > was around 0 - usually -0.7 seconds. The day I upgraded I started logging 
    > > differences of 12 seconds, every single day. I _vaguely_remember reading 
    > > something about this happening in recent kernels (when the CPU governor is 
    > > used) but no amount of Googling turns up any similar references.
    > > 
    > > Anyone else noticed this? I started noticing when the 11pm news started 
    > > getting about 10 seconds clipped off the front.

    > Is ntpd an option for you? It's what I use on all my pcs.

Does this persist across warm or cold boots?

12s/day is quite a high drift.  It's not -quite- more than ntpd can
handle, given good network conditions, but it could get close to that.
The problem is that if you drift too far, ntpd will declare that your
clock is hopeless and stop trying to correct the drift; at that point,
your clock will run away uncorrected unless something else (like your
daily ntpdate run) is keeping it at least a little disciplined.

I had this happen on one particular motherboard (unfortunately, on my
MBE) because its FSB spread spectrum was enabled; two other identical
motherboards with it disabled didn't do this.  It took months to
notice, because it was right on the hairy edge of failure and NTP
-almost- managed it for a long time (and the clock was getting reset
by a second or two every hour or so for a very long time).  Shortly
after fixing that, my backend ran -way- fast just once after boot
until I caught it (so fast the NTP PLL lost sync immediately and the
machine gained about 3 minutes in the first few hours).  That was
months ago, but today, entirely by coincidence, I've just detected
that my frontend/SBE (one of the identical motherboards) has suddenly
started running about 1.5s/hr fast; NTP is keeping it in check, but
I'll probably reboot after I've finished collecting some more
debugging info.  (Exactly what to check is a good question; this may
be some weird problem w/how the OS initialized the hardware.)  So
that's a couple of times in the six months or so that this combination
of motherboard & OS has mysteriously tried to run way fast.  (I've
been monitoring the situation ever since that first runaway, so I know
I haven't missed any in the meantime.)

(The motherboard is -not- using any sort of CPU speed changer, and
the OS is an old Breezy.)