NTP problem for virtual RHEL 4 server on VmWare

[Kenneth Holter]

Actually, most of our VM are running kernel version 2.6, and are causing
little or no problem with clock drift. But thank you for your useful input
on the subject - I'll take this into account the next time I come across a
drifting clock.

On 11/13/08, Eric Sisler <lbylnxgek at gmail.com> wrote:
>
> Kenneth Holter <kenneho.ndu at gmail.com> wrote:
>
> > What puzzles me is why this VM drifts so much, while most of our other
> VMs
> > doens't (at least not more than NTPd can handle).
>
> The Linux 2.6.x kernel is largely the culprit.  Are your other VMs 2.4
> kernels?  The kernel frequency HZ changed between the 2.4.x & 2.6.x
> kernels.  A snippet of a *much* earlier posting.  Unfortunately I
> don't have the name to give credit where it's due:
>
> "The 2.4 kernels have HZ set to 100, while 2.6 kernels have HZ set to
> 1000.  This value affects the frequency of clock interrupts that are
> generated, and those are per-CPU (that's why the problem gets worse
> when you go SMP).  This change in frequency is the main reason for the
> clock problems for Linux guests.  That is why your RH 7.3 works OK,
> while RHEL 4 doesn't.  Another drawback of this change in value of HZ
> is that 2.6 kernels will waste more CPU cycles than 2.4 kernels when
> idling.  Much more.  When you have several SMP Linux 2.6 guests, just
> to keep the kernels running will waste a lot of the available CPU time.
>
> One recommendation from VmWare is to recompile the 2.6 kernel with HZ
> reduced back to 100 (as it was in 2.4 kernels).  I haven't yet
> attempted this, but I heard reports that it works nicely.
>
> Apperently the behaviour of Windows is much closer to 2.4 Linux
> kernels.  Hence you don't see the problem that much there."
>
> As I mentioned earlier, adding a small patch to set the kernel clock
> HZ back to 100 & building custom RPMs does work, but can be time
> consuming if you have many servers to maintain. I *think* I have notes
> about the process somewhere if anyone's interested.
>
> -Eric
>
> --
> redhat-list mailing list
> unsubscribe mailto:redhat-list-request at redhat.com?subject=unsubscribe
> https://www.redhat.com/mailman/listinfo/redhat-list
>