Re: Server time wandering even with ntpd



Michael,

Michael Weber wrote:
This server is running on a SAN so there are no IDE drives. There are no
hard drives at all, in fact. All storage, except the CD, is through the
fibre channel card.

I have also seen the IDE issue messing with the clock, but I didnt think
it still had that effect when the only IDE device was a CD.

Basically the problem could be with any driver which is disabling interrupts
for too long so that timer tick interrupts get lost. However, IDE drivers
have been good candidates for this problem in the past, across a bunch of
different operating systems.

As others have already mentioned, it seems indeed more likely in your case
that the problem is due to SMI interrupts made by the BIOS. I don't see any
reason why SMI should cause lost timer ticks, but this should let the
system clock drift seriously even without ntpd, no matter which OS is
installed.

So you should really bother Dell with that problem and ask them to fix it.


Martin
--
Martin Burnicki

Meinberg Funkuhren
Bad Pyrmont
Germany
.



Relevant Pages

  • Re: why do PPP connections die with heavy disk usage?
    ... >> Because the IDE driver turns off interrupt processing completely ... > to/reading from the drive (that is, other interrupts aren't ignored ... > re-enabled interrupts (at the end of the IDE operation), ...
    (Debian-User)
  • Re: [RFC][PATCH] irq: remove IRQF_DISABLED
    ... Would you be willing to take such a patch? ... not enabling interrupts will not only make you ... historically resulted in even timer interrupts being lost because the IDE ... IDE IRQ handler assume that IRQ layer won't be ...
    (Linux-Kernel)
  • Re: NForce2 pseudoscience stability testing (2.6.0-test11)
    ... I didn't try yet was disabling the IDE controller, ... And my interrupts really are crowded, ... Then I bought the two SATA ...
    (Linux-Kernel)
  • Re: why do PPP connections die with heavy disk usage?
    ... >>IDE disk? ... If it turns off interrupt processing completely, how would, say, ... to/reading from the drive (that is, other interrupts aren't ignored ... >>than disk interrupts to make sure that serial data doesn't overrun ...
    (Debian-User)
  • Re: serial flow control appears broken
    ... More typical is something closer to 1 usec per 16-bit word. ... plus extra overhead for reading the IDE status register, ... I doubt that his system is using IDE PIO in the first place. ... I think that PIO transfers only have to be done with interrupts disabled on really old, ...
    (Linux-Kernel)