Re: recommended clock source

[Dor Laor <dlaor@redhat.com>]

On 08/03/2009 12:09 PM, Antoine Martin wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA512
>
>>>>>> It seems to be an unsettled issue, but, would any kind soul suggest
>>>>>> the current
>>>>>> best practice for setting the clock in Ubuntu Linux and Windows guests?
>>>>> For Linux the best source clock is the kvm pv clock (exist from 2.6.27
>>>>> and above).
>> # qemu-system-x86_64 -clock ?
>> Available alarm timers, in order of precedence:
>> dynticks
>> hpet
>> rtc
>> unix
>>
>> I see no "pv clock"...
>> Which one should I use then?
>> Did I miss a ./configure or .config option?
>>
>>> No, we were talking about different clocks.
>>> I was explaining the guest source clock while you wanted some info for
>>> the host-qemu clock.
> Hah, gotcha. You're talking about the guest kernel as in:
>   clocksource=[hpet|pit|tsc|acpi_pm|cyclone|scx200_hrt|kvm-clock]
>
> # cat /sys/devices/system/clocksource/clocksource0/current_clocksource
> kvm-clock
>
> So all good, thanks.
>
>>> You can use the default - dynticks. hpet and rtc might be good if you
>>> need a fine grain granularity on older<  2.6.24 host kernels.
> (I'm on dynticks)
> I guess dynticks reduces context switches on the host, but I still get
> hundreds per second on guests that are otherwise idle.
> How would I go about finding what makes them tick?

You can use the kvm_stat script to check the guest activity.

As for host timers, you can use sudo strace  -e trace=signal -c -p 
`pgrep qemu` to get the number of SIGALRM  calls.
You can reduce this number since there is a threshold in qemu for the 
dyntick.

My winXp guest does the following:
sudo strace  -c -p `pgrep qemu`
Process 14241 attached - interrupt to quit
^CProcess 14241 detached
% time     seconds  usecs/call     calls    errors syscall
------ ----------- ----------- --------- --------- ----------------
  82.46    0.089660          19      4801           select
  14.51    0.015775          33       480           futex
   1.70    0.001847           0      9511      4804 read
   0.72    0.000787           1       824           ioctl
   0.23    0.000246           0      2361           write
   0.19    0.000211           0      2463           timer_gettime
   0.13    0.000144           0      2361           rt_sigaction
   0.06    0.000066           0      2377           timer_settime
   0.00    0.000000           0         3           poll
   0.00    0.000000           0         3           writev
------ ----------- ----------- --------- --------- ----------------
100.00    0.108736                 25184      4804 total


>
> Thanks
> Antoine
>
>>
>> Cheers
>> Antoine
>>
>>>>> For windows, standard acpi HAL uses the rtc clock by default. As long as
>>>>> you use the -rtc-td-hack it won't drift.
>>>>>
>>>>> When the tsc is not stable on the host or the host cpu might get into
>>>>> deep sleep state (c2), you better use another source clock in the guest
>>>>> - for windows it should be the pmtimer (using the boot.ini).
>>>>>
>>>>>> Thanks!
>>>>>>
>>>>>> --
>>>>>> To unsubscribe from this list: send the line "unsubscribe kvm" in
>>>>>> the body of a message to majordomo@vger.kernel.org
>>>>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>>>> --
>>>>> To unsubscribe from this list: send the line "unsubscribe kvm" in
>>>>> the body of a message to majordomo@vger.kernel.org
>>>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.9 (GNU/Linux)
> Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
>
> iEYEAREKAAYFAkp2qT0ACgkQGK2zHPGK1rsSRQCfdGzkV8Gu+pTkZcnpXivXKQlt
> IrsAmwZAbfLzdyiWckoi80iwqc/k+0uA
> =Qt5o
> -----END PGP SIGNATURE-----
> --
> To unsubscribe from this list: send the line "unsubscribe kvm" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html