This is the mail archive of the systemtap@sourceware.org mailing list for the systemtap project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: gettimeofday_ns accuracy


On Fri, Sep 3, 2010 at 1:56 AM, Rob Donovan <rob@proivrc.com> wrote:
> Hi,
>
> Sorry for the delay...
>
> I think I've got a worse problem now with gettimeofday now.
>
> If you run a script for a long time, it seems that the gettimeofday_s()
> sometimes goes backwards by 1 second. And if you leave it running for 30
> minutes (or up to 2 hrs sometimes) then some probes are returning with the
> time 2-3 hrs ago.
>
> I have the following script below, and if you start it, it will start off
> with just a few errors of 1 second, but if you leave it running, it will get
> worse and worse.

Just speculating - does it still happen if you turn off power management?

Your output says the CPU is a 3.73GHz machine running at 1.4GHz, which
I think means you have frequency scaling happening. I'm wondering if
the larger differences are from CPUs getting completely shut down for
a while or something.

Can you print out the CPU ids in the output as well?


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]