> On 2023-12-13 12:02, Josh Moyer wrote:
>>> On 2023-12-13 02:51, Johnny Billquist wrote:
>>>> On 2023-12-13 11:46, Josh Moyer wrote:
>>>> port-vax/56383: KA630 slow clock drift For those doing compiles or
>>>> other high load jobs (and not just on KA630), keep an eye on your
>>>> clock drift. Clock accuracy seems significantly improved in
>>>> 10.0_RC1 - although it's still not perfect. Make sure you aren't
>>>> wasting cycles and masking the problem by running ntpd.
>>>
>>> Are you saying that we loose interrupts? Basically, we are running
>>> for longer periods with interrupts blocked? That sounds both like a
>>> serious bug, and something that should be fairly easy to detect, no?
>>
>> I don't know the root cause, but interrupts seem possibly likely.
>> Would that also lead to the terrible accuracy of top and the cpu usage
>> statistics?
>
> Basically, if the drift gets worse under load it must be lost interrupts.
> There can't be any other way/reason.
>
> I don't know what the "terrible accuracy" thing is.
See attached PNG. The values in the CPU and WCPU columns don't sum up to
the percentages in the CPU states line at the top of the display. And
the sum of the CPU status != 100. This was happening in 9, too, but it
might be worse now.
Kind regards,
_____
| * * | Josh Moyer (he/him) <JMoyer%NODOMAIN.NET@localhost>
|*(*)*| http://jmoyer.nodomain.net/
\ - / http://www.nodomain.net/
\//
Love, Responsibility, Justice
Liebe, Verantwortung, Gerechtigkeit
Please don't eat the animals.
Thanks.
Attachment:
top_10.0_RC1.png
Description: top_10.0_RC1.png
